I was laid off a few weeks ago. Then I learned my dad’s carpet cleaning company has been struggling, so I started learning about Google Local Service Ads to help him out. THEN I learned his old Windows desktop CRM was discontinued and one OS update away from no longer working. I wanted a way to track every dollar spent to determine return on ad spend, but the Google LSA dashboard left a lot to be desired. And existing SaaS CRMs didn’t have great integrations to Google Ads API. So I built Crewbird, a Field Service Management and CRM for service based businesses with Google LSA integration where my dad is quickly approaching 3x ROAS!
I’m still looking for work in the mean time but I’m proud so far of what I’ve achieved.
I’m a product manager fascinated by the skills of good salespeople. I also like Duolingo. so I’ve been building a “Duolingo for sales” to try to teach myself more. it’s started to morph a bit into an eval benchmark for sales agents, so classic cognitive feature
creep.
making it feel fun versus a basic quiz workflow is the hardest part.
As a self hoster whos paranoid about backups. I wanted to cheat a bit and use s3. With growing resentment on big tech, who I worked for earlier, I'm now building tools that empower myself and other users and decrease reliance of big tech. Arkel is an object store I've been building and testing for the past few months. https://arkeldata.com
A singing training app for bad singers. You sing the pitch line and get a score. Let's you sing to any original song by listening along and not playing the song itself.
Recently got some good feedback on HN https://news.ycombinator.com/item?id=50195488 and am following up with Linux and Windows versions and using Fable 5.1 to internationalise the app and landing page which explains the app.
I'm working on Zustri (https://zustri.com), booking and scheduling for therapists and salons. My girlfriend is a therapist and didn't like any of the booking/scheduling software she tried, so I built her one.
I finished the basics a long while ago. Since then I've added an AI agent, a public booking page, and support for multiple staff. Hopefully, I'm launching soon. Currently finishing final touches, gathering feedback and trying to figure out how to market this
I work on a suite of open source public transit software projects, collectively called OneBusAway (or OBA). The software is used by millions of transit riders every day around the world, including in New York City, the Seattle area, San Diego, and Washington, D.C.
University of California San Diego has deployed our Wayfinder and Waystation web apps for their 35,000+ students!
We're getting ready to launch our Vehicle Positions software, which replaces expensive, proprietary vehicle location systems with free, open source software that runs on commodity smartphones.
Our next-generation server software, Maglev, is adding support for newer transit data formats, like GTFS-Flex so that we can show demand responsive transit systems on the map.
We are always looking for more developers, biz dev, ux designers, product managers, and more to help out!
I got tired of always having to update my wife, and my family members résumés. So I created https://theresumechef.com
It's private,free to use, no pay walls or monthly subscriptions.It has10 resume templates to choose from to get you started.The final resume can be exported as a PDF, or stored on Google Drive or Dropbox in JSON format. There are still a few issues, but I've learned a ton by just creating and launching this project. I have some exciting new product ideas, but I'm currently waiting on my Dutch KvK number to accept online payments with Stripe.
When I was a kid I loved a game called "Stars!" from 1995. I cloned the core and I'm also adding my own touch and lots of high sci-fi concepts. Already fun to play, early access coming soon: https://store.steampowered.com/app/5482440/Asteria_4X
For the last 3 years, with a small team has been building https://kube-dc.com/ - a platform for building enterprise grade private and public clouds. Its like VMware + OpenShift + Railway in a single product: completely multi-tenant, built entirely on the K8s API (we built over 46 custom controllers in Go, with UI in Radix UI), with managed services like nested K8s, Postgres, Kafka, Valkey, and GitHub runners. Not the charts :) real services. Also we can do vlans BGP, shared GPU and all fancy stuff required.
Now we don't deploy this to private companies except some gov projects. But we are engaging existing IaaS and dedicated servers providers. Later hope we would fight with Broadcom and IBM :)
We previously ran a services company that built over 100 private clouds, then decided to turn that experience into a product with an aggressive level of automation to let DevOps, SRE, and platform teams hit a 1:300 ratio per developer (now dev + AI agent teams).
We are currently deployed across 12 datacenters and raising our seed round this month.
I’ve been working on Beacon, which just got approved on the App Store! It’s a mobile app to instantly answer “which friend is free to talk right now?” and summon them in a single tap.
Send your beacon to a group, and everyone gets a gentle ping. The first person who answers gets connected for a 1-1 call, and for everyone else the signal drops silently. It does not log any missed calls so there’s neither pressure to answer nor pressure to return calls.
Overall it works really well! It’s best if you're able to join with 4+ friends/family you’d like to chat with more. I have dozens of connections in the app now and it feels like absolute magic to me. I tap the button to send my beacon to everyone and get a random friend on the line.
This is really cool! I'm very excited to try it - nice to see it available on Android also. As someone who has messed around with building relationship-focused social apps, I'm curious if there were any other concepts you tried, or other apps you like? What did you think of BeReal for example?
Just like the last time I posted in this thread, I’m working on The Marvelous Brass Chessplaying Automaton, a chess database/playing/opening training app for Apple platforms:
The original goal was an alternative to ChessBase, but with an Apple-native GUI and iCloud sync between devices, and as an improved alternative to tools like chessbook.com for operating repertoire management and spaced repetition opening memorisation. In recent months I have focused more on the chess-playing parts of the app, trying to implement human-like computer opponents at all levels, using a transformer-based model (similar to the one described in the Maia 3 paper) combined with a very small search.
Recently I’ve also been working on chess variants. Since a while, the companion Apple Watch app plays 5x5 chess. More recently, I added a 10x10 chess as an Easter egg game inside the app, and I’m now working on hexagonal chess. With the current state of the art in computer chess, it’s almost trivial to implement vastly superhuman strength in all these variants, but implementing weaker levels that are enjoyable opponents is interesting and challenging. Unlike in normal chess, there are no big databases of human games to train on.
I have been playing with benchmarking agents in a research loop. The experiment leans on using these models to understand and explore builds/gameplay through reasoning game AI over traditional reinforcement learning.
Also, creating interactive ways to view the experiment and visualizating the results.
I've been deep in the weeds on two wildly unrelated projects.
Project #1: I've been working on migrating an application [1] from GCP (where I pay $200/mo to host it) to a physical server (PowerEdge R7625) I've racked in a colo near me, running on my own IPs/ASN [2] and all that jazz. I've been dragging my feet on the this as there's just a lot of moving pieces, but I got a good chunk of the last few hard bits done today.
Project #2: Building a DIY bright-ass lamp, like Brighter [3] or Rhythm [4], but for a few hundred bucks instead of $1-2k. I've specced out the parts, designed the hardware, and built the firmware. Later today I'll do my first LED test against a bench supply, to get some hard numbers for if my cooling solution (a chunky aluminum heatsink and some Noctua PC fans) is sufficient for the full 500+ W final version.
As far as curiosities go, I've been doing a lot of Go (the board game, not the language) puzzles, and reading about sailing (mostly the book from the Hundredrabbits folk [5])
https://santawho.com A funny holiday game you play in person with the whole family on just one phone (the parent interviews everyone for jobs on the North Pole and surfaces inner insights about what everyone might enjoy from Santa). It’s for any age, private, ads free, just pure holiday spirit.
https://www.dreamlist.com - collaborative shopping lists and online wishlists and gift drive platform. Again ads-free, private, stack multiple lists into groups of lists to share on social media or with grandparents.
A number of other ones about to launch as well. All little tech, not big tech - private by default, no ads, no feeds or media, just surfacing what is needed.
I’ve been working on my inertial navigation library. It actually beats ArduPilot’s navigation subsystem in my benchmarks. I’ve been working on these algorithms since my PhD thesis and now I am releasing the stuff as open source embedded C and Python library:
I'm working on a better way to design longer scrolling pages with multiple sections, while preserving the visual look/feel and drastically speeding up the experience.
As a disclaimer I cut 32 seconds of generation time out of that video, but still I'm quite happy with the speed. Not launched yet, but it's been fun to work on / wanted to show off progress :)
Rebuilding my ERP to work more like the model of https://plaintextaccounting.org, where I expanded the idea of `double entry accounting` to `double entry trade between parties`, plus several other ideas for what I see a way to collapse a lot of complications that are typically for ERP systems.
I make a special `textual DSL` where I model business scenarios like `Make a sale that include 3 invoices : 2 to different cities in country A in currency USD and other in country B in currency EUR` + all the intermediate steps like transportation, manufacturing, etc.
I`m now in the woods with a simplification for the price engine, I dislike much how complicated is normally, and I think I have good enough solution for most cases that generate without dynamic calculations.
I have been in enterprise for couple decades now, learnt and experienced the challenges/gaps between the specs vs implementation, the asks vs the real needs in the space of identity, security and access
Product: webfrolic.com
Demo: demo.webfrolic.com
What it is?
Consider Okta (Workforce identity)+ Auth0 (Customer Identity), Authzed (authorization) + ServiceNow (IT service Desk) + Zendesk (Customer Support Desk) architected together
Being an engineer there is this inherent habit of prioritizing correctness over marketing so I would not oversell it, though its probably the most secure and complete thing that could be built from identity perspective
Status (WIP):
1. I am still debating/settling on how to incorporate the AI Agents needs into the architecture, big focus on avoiding the creation of security or performance foot guns where administrators leave room open for abuse with configuration despite all this attempt to build secure system
This one I (mostly) did without LLMs, but the other project is vibe coded, but I'm enjoying it. An RTS made by Claude in Unity: https://github.com/SiGe11/StarForge2
A slow, realtime strategy game similar to Neptunes Pride and Subterfuge.
Each game tick is an hour and games can last over a week. I got hooked on both Subterfuge and Neptunes Pride and always wished there were some more titles in this genre.
SuperBlocker is a native blocker that works on iPhone and iPad using Apple's latest URL filtering technology that blocks URLs without revealing it to Apple's relays or SuperBlocker's servers.
The original JSON Query app on the Mac App Store. Store, organize and query all of your JSON with jq or natural language. All of your data stays local so sensitive JSON can stay on your machine.
I've started maintaining my own Firefox fork that:
- removes the stuff that I don't care about (Firefox sync, VPN, etc.)
- fixes a lot of the awful UX/UI choices that the (frankly) incompetent product team at Mozilla refuse to
- actually adds stuff that I do like integrated mail and calendar, a document library system (basically Notion), native ad-blocking, terminal, task-tracking like Trello, time-tracking like Harvest, etc.
- has integrated AI (Apple intelligence for MacOS for the easy stuff and Claude for the hard stuff
It means that instead of having half a dozen applications open, I can just work within my Firefox fork. Because it's all deeply integrated, I can do wild stuff like:
- tell my AI to search the web on a schedule and send me a custom daily brief with my own email
- maintain a comprehensive memory of all of the work that I do, across half a dozen clients, and be able to query it easily
- pull analytics from multiple web dashboards and format them into something cohesive
- program the web with a node-based infinite canvas that has live websites as part of the workflow
- manage my tasks across multiple domains (for example, it will see that I sent an email to a client and automatically update a task that says that I need to do that)
It's incredibly powerful and it has made me much, much more efficient.
A video game in which players can build their own worlds. The brains of the creatures in the video game are a small language model + harness that runs in your browser: http://id-le.com
I've always been interested in health, longevity, and biohacking.
I'm Canadian but I moved to USA to work for Microsoft. One thing I liked about the USA was how easy it was to order bloodwork/lab testing. My original idea was to bring that experience to Canada, but since starting I've found other ways to improve the patient healthcare experience in Canada too.
ColliePWA[1] a mobile interface to continue working with your agents when you're on the. Trying to meet devs where they work on the desktop in multiplexers (herdr/tmux/zellij). Remote access can be set up with any front door like tailscale.
Still not sure where this is going as the space is super crowded and moving fast, but it's solving a lot of my own problems and doesn't rely on third parties to run.
I needed a code-first approach for making installations and interactive art. MaxMsp gets messy with very big projects. Python didn't really work well for music making and Tidal/Pi/Sclang are only meant for sound. So I needed something that has bpm synchronized threads, but can draw and do anything else like file IO, networking, hardware control etc. It works quite well now.
As many of you, I have a few subscriptions to Claude and Codex (just me??). It's a daily occurrence to use 2-3 accounts for the same conversation (whenever the limit stops me).
What I need is something to switch the account where the usage is taken from, not the configuration.
I'm not sure providers would shoot themselves in the foot by not allowing you to buy more usage. It seems like both OAI and Anthropic could use all the subs they can get.
A suite of high-dopamine games which use smart exercise bikes as the control interface.
When playing Zwift, Rouvy, etc, I find that I get too competitive and ruin longer workouts by trying to keep up with or sprint past people who are much faster than me over long distances.
This is bad for building stamina and endurance, but I get bored, and it's hard to fight boredom and physical strain both at once.
I do not get bored sitting in front of Balatro, Raccoin, Tetris - anything with bright colours and satisfying noises and combos.
The first finished prototype game is very simple: you sort coloured marbles. Higher watt output from the bike means marbles are blown, by wind, upwards as they roll across the screen. Lower watt output means they fall. Steady watt output means they move horizontally. Marble goes into correct hole -> satisfying noise. It rewards steady watt output over time, with minor adjustments, which is what I'm trying to work on right now.
I was pretty amazed by how engaging I found it (total iPad baby brain). The miles really fly by - I get tired before I get bored.
I think that next I'll either do a tower defence game, or a less-dopamine-rewarding game in the style of Kerbal Space Program. Taking your actual weight and watt output, putting a bike with wings into a physics engine, and the gameplay loop being unlocking increasingly large airplane pieces, and designing - then flying - increasingly ridiculous airborne bicycles seems like a compelling idea, though less good for fitness.
The control surface being so limited - essentially one measurement which can either go up or down - is a really fun constraint for creativity.
There's something in this idea. you can actually tie it to a smart watch (Amazfit allow for custom apps trivially for such uses), and you have a bigger pool.
Thanks, yes, I agree - it really doesn't "feel like" exercise when you're being bombarded by dopamine rewards. Despite this, most "exercise games" leave me with the same feeling of boredom / "fake fun" I got from playing educational games on the computers in the IT room at school.
I'd love to try a similar approach to gamifying exercise on a smart watch. That's an even more fun problem, because (when running, for example) you don't even have a constantly-present screen to communicate dopamine to the player with. Perhaps some watches wouldn't even be able to provide sound feedback? I've never owned one. It's a really interesting constraint for game design.
I'm working on an iPhone app called PhotoGather. https://apps.apple.com/us/app/photogather/id6738232616
I've got 40k+ photos on my iphone, and wanted a better way to interact with them. The basic idea I had was to virtually stitch all the photos together into a gigapixel image, and navigate that image with pinch and zoom. On import, the app hierarchically groups photos based on date and similarity. The user can go back and edit the groups, change the layout, and adjust the size of photos if they choose.
The interesting I noticed is that in the process of curating and organizing my photos, I stumble across memories and moments that I had completely forgot.
Development Challenge: The app does two interesting things underneath the hood... First, given an arbitrary number of thumbnails with different aspect ratios, it finds the best sized rectangle that completely fits all the photos. This allows me to recursively stitch groups together and always output a rectangular thumbnail for the group. Second, the pinch/zoom UI uses these nested thumbnails directly without slicing/dicing to fixed sized tiles.
Still working on ShopSpec, a parametric deterministic furniture builder. Trying to find time to put more actual builds in the gallery - have a few projects that are in progress while continuing to build out the tool and iterate where I find personally useful.
Hunting bugs in the Python ecosystem using fuzzers and AI tools I develop. If you have a Python package or extension you'd like analyzed, just let me know.
Meli: A spending tracker for Mac designed with privacy in mind. Turns credit card alert emails into spending insights and budgets. I was in security for 15+ years and I built it because I didn't want to use Plaid or other aggregators and wanted my financial data locally on my device and not in the cloud. Spreadsheets required a lot of work to maintain so I wanted some sort of automation. Connects directly to your email (reads only alerts from supported senders) and all parsing happens on your device.
Works with seven major US banks (AMEX, Capital One, Chase, Citi, Discover, Bank of America and Wells Fargo) plus PayPal and Venmo.
Does this list cover most of your spending? If not, which banks or cards would you need?
Like 2h ago I was doing some research on that topic. I’ve heard about open banking a while back and thought: Oh that’s easy. I’ll just get an api key from my bank or “oath” into it.
Sadly it’s not that easy.
Unfortunately it's not easy. In the US specifically banks don't generally offer APIs or keys to access them to individuals and you have to go through an aggregator like Plaid (which is what I want to avoid).
However, most banks do send alert emails when something is charged to a credit/debit card. Meli connects to your inbox read-only via OAuth (Google and Microsoft) or IMAP for other providers and reads those emails on your Mac.
Of course there are limitations to this approach, you only get access to the data that is available in those emails like merchant, amount, date and account.
Which banks are you interested in? Are you based in the US?
I’ve been working since the start of the year on a minimalist web reader (https://papercove.app) that I use every day to follow RSS feeds, newsletters and read-later, and that helps keep things from piling up.
I thought I’d get where I wanted much faster, but it turns out I ran into quite a few challenges: it’s quick to build something simple, but it takes much longer to get a reader that’s robust and pleasant to use (extraction, parsing, security, infrastructure, etc.).
On the tech side, I’m using my usual stack, Laravel and Livewire, with some Python for extraction and Go for the email infrastructure. This month I’m working on the beta release, since several people around me were already using it, so I’d like to gather more feedback.
That is a great landing page! I can relate on the velocity side, it can be quick to build a feature, but it takes time to get everything feeling right and usable.
Serverless Wordpress + MariaDB without requiring plugins or any modifications to customer code: https://www.agiler.io
The serverless MariaDB is important because it can scale to zero, so you have dev sites that cost less than $0.50/mo.
Also added MCP and CLI tooling that exposes all of the infrastructure + WP CLI + DB so agents can develop WP sites... which is nice because you can still use the WP admin interface, while using AI to develop the site, debug it, change pages and posts, etc.
A proportional width font terminal. "i" and "w" shouldn't take up the same amount of horizontal space. We used to have to do that because computers were slow and we counted cycles and every bits mattered. We live in a land of plenty now though so we can spend cycles doing math so that ls -l/top/etc table output is still aligned but "i" and "w" can take up the appropriate amount of horizontal space.
I'm building an open-source "spatial workspace for directing AI coding agents like an engineering director".
I got overwhelmed by walls of text of parallel Claude Code sessions. This limited the sessions I could steer in parallel to maybe 5 or 6. I knew there had to be a better way, and I found one: A canvas per project. Work laid out as cards. An agent in a worktree per card. Chat condensed to WhatsApp-style back and forth, not a litany of tool calls. Result is usually a narrated demo video that I can comment on by voice to have the agent go another round. Remote-control from my phone. Remote workers to scale beyond my machine. Everyone I know who tried it has stuck with it so far.
I’m still working on https://feedsync.net, a tool to push RSS feed updates to Discord.
It’s been getting a decent number of sign ups despite me not really posting it anywhere or doing any marketing.
Ultimately a fun project, and useful for myself and a few friends, so I’m happy others are even interested in trying it.
I’ve started to regret using Phoenix LiveView, as it’s only deployed in one region, so high latency far from there. I’m exploring a SPA as a better alternative.
Otherwise, I’ve been playing around with a Lua engine to build feeds for websites that don’t support them, using AI to build the Lua as a once-off. It works reasonably well, but isn’t ready for prime time just yet.
Noticed how well Claude works for transcribing old German cursive, especially for broad hits in non-indexed 17th century parish records. Motivated me to pick up Ancestry again and helped me quite a bit in following some lines further back.
Then also HNowledge[1], a free iOS app for browsing the HN stories and comments I've upvoted. Made it to fit the way I browse HN, which is upvoting mainly what I find interesting and may want to come back to. The app lets you browse by date, domain or score, add notes and export. Apart from syncing your upvotes from HN, everything stays local without any tracking.
Wanes[2], an Apple Health logging interface for tracking how long symptoms last. I tend to have headaches quite often and the way Apple lets you enter that data is pretty limited, so I created an app which starts a Live Activity for one or more symptoms using widgets, Control Center or Siri. You can change the severity rating while it runs and stop it from wherever you are on your phone (Home Screen, Lock Screen, Dynamic Island, Siri).
I'm making a replacement motherboard for the Nokia 1100 that runs Linux and supports LTE. I have the software working well on a MangoPi dev board and I've designed two PCBs: one "test fit" board that just breaks out the phone keypad/LCD to pin headers, and the actual board with everything on it. Now I just need to order the PCBA which is super expensive :)
It uses an allwinner T113-S3 SoC and the Quectel EC25 LTE module, and I've made sure voice calling works over LTE, at least on T-Mobile in the US. (this is tricky bc of carrier rules)
Building a soccer game based on Mike Tyson's Punch Out, just to see if I can create some method to see if an agent can iterate on making something more fun. Obviously that depends on the definition of fun (which is already a can of worms), but it's a fun way to use a different side of the brain.
Because I can. Well, let's be honest: because LLMs can. There's nothing particularly novel about this. It's just wiring a bunch of libraries together. Certainly would have been time consuming, but not difficult, strictly speaking, since all the heavy lifting is done by the libraries utilised.
Likely going to migrate https://breaka.club HUDs and editor UI to React since Godot control nodes are by no means impossible to work with, but somewhat cumbersome.
EDIT: Oh, will be open source like a lot for Breaka Club's other libraries very soon. Still optimising.
I built a new abstraction layer for code (https:// www.archme.io), as I believe it's no longer the right medium for understanding and collaboration in the era of AI and 10k LOC PRs.
It scans your codebase, turns it into permanent c4 & sequence diagrams and keeps them up to date on each commit. Then, instead of reviewing a 10k line coding PR, you review a 30 line change to sequence and C4 diagrams.
npm i -g @archme/cli && archme init to run it on your own computer, code never leaves your machine.
I have a PAL-2 [1] (a modern KIM-1 re-creation) and started learning 6502 by entering programs from The First Book of KIM [2].
I wanted to be able to compile/link the assembly from the book so leaned on Claude to set up a 6502 dev environment for my Mac. I put everything needed (and instructions) on GitHub [3].
(Claude also has been helpful in re-formatting the 6502 code from The First Book of KIM into a more modern style compatible with modern assemblers. A few of these modernized programs I added to the GitHub repo above.)
As it happens, I am also fixing/formatting The First Book of KIM by recreating it in a modern page-layout app (Affinity Publisher). The PDF scans that are out there are not pretty, have OCR errors (and the book itself has mistakes). Hopefully I'll have a new version of the book/PDF out on the internet by the year's end.
Repair of a Collins 651S-1 receiver. The 65-95 mhz vco module is bad, but if you over damp the pll loop filter it does eventually lock and receive properly. I found a similar but not identical replacement from a different model receiver. It lacks one of the two DC tracking signal outputs but I added an op amp based circuit to simulate the missing signal.
The loop is stable and responsive when receiving from 17 to 29.9999 Mhz, but not much below that... Almost got it
I'm still working on making JustFax Online[1] more profitable. Added a bunch of features recently, as well as invisible improvements in fax routing, and admin stuff for myself to better analyze the traffic and patterns.
Other than this, I'm playing with vibe-coding stuff I always wanted to do but never had the time/skills to do. For example, I built a TUI on top of beancount that is tailored to my needs and my business[2]. I also completely got rid of self-hosted plausible and umami analytics, and replaced them with vibe-coded built-in-rust (c) analytics tool that relies on both js-beacon and streaming of caddy access.log. Will probably publish about it in my blog soon, so subscribe to the RSS if you want.
Lastly, I'm messing with self-hosting. I completely ditched Github and Gitlab for self hosted forgejo + actions runner. I set up proper backup plan for everything I care (documents, immich library, my SaaS DB, and the forgejo repositories).
https://skymap.rulkens.com
A browser based open source universe explorer. I'm in between jobs currently and finally took time to work on my passion project. It started with just visualizing some galaxy surveys, but it got a little out of hand and now includes everything from hi-res tiled earth/mars, Gaia star field (16M stars), a milky way renderer and much more.
It's so amazing to have the ability to gather and parse publicly available scientific data and actually visualize it in real-time. It's one of those projects I'm learning a lot from and can keep going with (the universe is a big place :)). As a software developer for 20+ years, its been mind-blowing and often also frustrating to have Claude generate code that I deem good enough. About half of my time has actually been spent on refactoring, building quite some skills in guiding LLMs in the process. I guess that part still hasn't changed in a growing codebase.
I’ve been working on an accounting agent factory (https://www.translucent.io) .
Building a harness and infra that can process books with thousands of invoices, transactions, etc. is quite rewarding and challenging. In particular the long-running nature of things.
As a heavy LLM user in my daily work, I've been working on optimizing agent memory on complex projects. That resulted in this simple Rust-based tool called Wayfinder: https://github.com/zaterka/wayfinder
I've been trying to improve agent knowledge over a codebase, and existing approaches weren't giving the result I wanted, and I wanted to keep it simple. In the end, I added a semantic layer on top of an LLM Wiki and it saved 30% of token usage in multiple benchmarks I ran. It also adds some nice features to deal with concurring agents updating the shared memory, to ensure all information there is reliable and fresh.
It's published to PyPi, so an "uv tool install wayfinder-memory" already does all the work, and it's compatible with almost all Harnesses.
APIKulture. An alternative to Postman/Insomnia/Bruno.
To be honest, I started this because Postman and Insomnia became forced-corporate-cloud not usable for me anymore, and Bruno lacked a feature I considered fundamental for my workflow.
The feature is to remember the JSONPath expressions between requests, a feature that Bruno has implemented already.
But I created this thing and I'm using it daily since then. C++ and wxWidgets, it uses 40MB of RAM, and it is super fast, instead of the 400MB used by the above mentioned programs.
1. Broadcast (https://sendbroadcast.net) - Self hosted email marketing software. Been working on this since 2024 and positive revenue. Releasing v3 this week! :)
It's a free job site with some resume tools bolted on. Lots of people from all over the world are using and that feels great. It has 2.7M+ jobs and automatically removes any job older than 7 days. We never allow reposts because, frankly, the job market is a nightmare and any job that has been up for 7+ days almost certainly has 1000+ applicants already. You're just feeding your resume into a wood chipper.
1. FormBeep(https://formbeep.com) at $38 MRR now - A form to Whatsapp tool. Stack is mainly CF workers with Clerk Auth.
2. SlashLink(https://slashlink.app) $0 - A simple pay as you go amazon link localizer. Great for affiliate folks and a cheaper alternative to Genius. Stack is same, CF workers with SQLite(D1)
3. StackStats(https://stackstats.app) $791 - A better advanced analytics app for Substack writers, fully local. I have been writing on Substack for a while now and their analytics suck!
I try to 'vibe code' a product every 2 weeks. Most of them go unpublished, I just build it for the sake of building it. Most recent project is Unmute Party (https://unmuteparty.com) - multiplayer party games for remote teams, played in the browser during a Zoom or Teams call. The server is EU-only for now, and I plan to add US servers next week. Most of it is free, the paid tiers aren't even working yet. The paid tiers can be unlocked by using code HNWHATAREYOUWORKINGON.
Apart from that, running a travel agency I'd never be able to without agents (they manage my ads, social media, website content etc.). Started with boa.ba and it went quite well, now preparing dobosnia.com
Still cracking on with https://whistle-enterprise.com and making improvements to the document generation model. I'm ditching my old trusty Phi4 model and testing out some other small models to fine tune, something about Phi4 just seemed to work quite well for what it was so it's a hard contender to beat for midrange laptop CPU inference at a decent rate and quality.
You can connect Claude desktop to databases using MCP servers, but Claude desktop is not great at displaying large volumes of data. A custom client can do a much better job.
It is a full-featured reporting tool, and can be used without AI. If you do use AI it has a semantic layer to help AI understand your table and column names, and relationships between tables.
Still plugging away on Flopper to build a holistic view of the GPU market including providers, rental prices, vendor information and mostly - specs for GPUs.
Column view for hacker news with resizable fonts. It’s just a fun hobby thing for me. I’m not in software, but I can do hobby level programming and that gets leveled up by codex.
Working on / dog fooding my graph based recipe app[0].
There are more things I'd like to do with the graph concept, like linking two user's recipe trees if they were imported via the same URL. Really trying to to avoid features that I think would be cool that don't provide any functionality.
There are still a few features I'd like to implement that are more in the vein of nice to have in any recipe application, like nested recipes, temperature conversions, volumetric ingredients amounts to mass conversions, etc.
But I've been using it for new recipes and converting my old recipes I have written down in the margins to this format and I've been happy with how usable its been in these early stages.
I've been using the recent MIT weight-space programming techniques to work with neural quantum states, which are neural networks that represent a quantum wavefunction. I was able to produce a bunch of novel networks for a few hundred dollars of cloud GPU, which would have probably cost tens of thousands to train in the traditional manner.
I'm still working on my own shell based AI tool. This month I am testing and reworking it to try and get useful results with really small models, like 2B level. Goal is to get help debugging Linux issues when offline with these small models that can run anywhere. Then also being able to access cloud based models to do harder stuff once online. all while feeling like a CLI tool.
Artists make users jump through hoops to like, comment, repost etc.. before giving them access to the download - CrateDiggr highlights the free downloads in the Soundcloud UI and clicks through the download gates for users.
Been a fun passion project, check it out if you download tracks from Soundcloud!
Still changing along on Incredible (https://www.incredible.rs/) a Rust Text User Interface (TUI) Framework for the 2nd Quarter of the 21st Century.
It is New, Experimental, Modular, Declarative, Event Driven and Reactive. It builds cross-platform apps that can be compiled for the terminal (macOs, Windows, Linux), a standalone native window (macOS & Windows 11), or the browser (via WASM) from one unified code base.
Curious if this project will ever be 90% done.. ;)
Lipik, a free open-source dictation app for Mac (https://github.com/Gourav2609/lipik). Hold fn, talk, let go, and clean text lands in any app: ums and self-corrections removed, spoken punctuation turned into real punctuation. Speech-to-text and clean-up both run on the Mac. On my 106 English test clips, lines that came out fully right went from 28% to 76% with clean-up on. Right now I'm working on Hinglish (Hindi + English in Roman letters). It gets 7 of 20 test lines fully right, so there's a long way to go.
https://www.haptrix.com - An Core Haptics editor, with the release of the iPhone Duo, the editor finally makes sense on device, so I've been working on the new UI.
https://www.phonique.com - An AVAudioEngine/PHASE Audio graph editor, it's easy to connect nodes together visually, then export the wired graph as Swift
HostsTray, tray utility to manage your hosts file, that also has a reverse HTTPS proxy builtin, that installs a self-signed CA and generates certs on-the-fly.
Also it forces your browser to "flush" their DNS by killing the keep-alive connections that for example Chrome keeps open in background. And it can be controlled by your local coding agent through a simple API (which also serves the skill for it)
The work mainly stems from a belief that Nvidia gpus' architecture are a bit subpar for machine learning and that tpu could be a viable alternative but are still rather undocumented.
Building the web-embeddable legal IDE for transactional AI-native law firms. Docx, PDF and Markdown in the browser with redlining, go-to definition and more. In Rust, of course.
For the first time in along while, I'm not actually working on anything. I do some vibe coding for fun.
I'm kinda leaning towards vibe coding my own GUI coding agent. I currently use Pi and recently have been playing with Herdr and Paseo, but none of these feels right to me. These are amazing tools, but ai think we could do better. But I'm very hesitant - I know the only way to "succeed" in a project like this is to write (vibe code) hundreds of thousands of lines of code and it's kinda depressing.
Few apps I have originally mainly done for myself; Vaava (https://vaava.app) , been tweaking the e2ee data sync between connected apps, and the webrtc based baby monitor feature stability. Crazy how easily Astra / opus 55 level models create quite complex ad-hoc reproducible (almost automatable) test scenarios with 2 phones, laptop and other tools / peripherals they know they can use, for example for the baby monitor feature Astra did a setup for me which i can reproduce to test that the video + audio works fine between 2 phones for x time.
Also trying to replace own Duolingo usage with Fika Lingo (https://fikalingo.com) basic idea was to first repro the practice of matching words. There is now few other types of language / vocabulary practices. The practice generation works on-device with local LLM, you can give the topic yourself like "Words that relate to Hacker News". Downside at the moment is that English needs to be the counterpart of the practices. Cant guarantee good enough quality with something like germany <-> italy practices. I'm trying to do finetuning with LFM models to see if it would be feasible to improve the local practice generation.
I'm working on Eigakan Guide (https://eigakanguide.jp/en), a cinema showtime aggregator for Japan. Every popular cinema site in Japan only allows for very few filters, making it hard to find exactly what I was looking for. So instead I built a site where it's possible to filter by area, cinema, date, time, features like IMAX, English audio, English subtitles, and more. This allows for searching for stuff like IMAX showtimes with English audio on Friday after 7PM.
Since the last time I posted on HN I've added 100s of cinemas, user features like being able to save search filters, movies, and specific showtimes. Also a ChatGPT plugin which allows for searching for movie showtimes using natural language directly inside of ChatGPT (https://chatgpt.com/plugins/plugin_asdk_app_6abfdc298c208191...).
Finally I've added AI search to the site as well, allowing for natural language search with queries like "English audio showtimes near Shinjuku tomorrow" and so on.
I would really appreciate feedback for things to add, things that are hard to understand, and anything else!
Getting https://streamish.github.io/ ready to launch, a music server compatible with multiple 3rd party mobile apps and particularly Synology and QNAP. Want to do Plex next but I suspect that's a dead-end since I think you can't directly sign in you have to do it through their website.
Now putting it to personal use and keeping an eye out for issues and rough bits to smooth over.
Working on a modern implementation of a torrent client [1], it's already pretty usable for most scenarios. Provides both a daemon + web UI out of the box as well as an embeddable engine for any Rust application.
Working on ZShot [1]. I'm about a year in on a local-only browser tuned specifically for CLI and API usage - kinda filling the gap for "geez, I sure wish I had a browser, but not one to look at". You need a webpage as PNG, or an MP4, or a website as a WACZ or SQLite database. There's a Mac app to help with discoverability of options, but I wanted for my local tooling and building services on top of it.
Along the way I ended up adding an internal agent framework to satisfy stuff like "log in, click the button, and then take a screenshot", out of which fell stuff like automatically generating Playwright tests from that. It even runs the Playwright test suite and Puppeteer test suites ( a bit faster, and for a lot less memory ). I kinda love this bit because I can layer in multiple models that will bounce control of the web session between each other, to do stuff like let a Typesafe model choose to escalate to Opus when it needs higher minded behavior. It's fun to watch it happen.
I've mentioned it a few times around here, I'll probably do a Show HN soon when I feel like it's ready for that. Hopefully soon.
the plot: Programming languages existed as an abstraction for humans. As writing code becomes cheaper, we can write it for each architecture as long as the specification is done well. The costs of running will matter more in relation. So don’t make your AI write rust, but assembly
An opensource personal expense tracker, I was a happy user from an offline expense tracker which wiped my data when playstore decided auto-updating it,
I'm at the point where I'm finally actively using it to register my activity, it is straightforward to register tickets by sending them to a telegram bot, then, reconcile those by uploading my bank statement.
I'm trying to get the lowest possible usage friction so that I keep using it.
So far the better built pieces related to the data ingestion. Once this is good enough I'll build reports + anything coming to my head as a useful piece.
I'm leveraging AWS textract + bedrock + bda to parse bank statements. I expected this would have been easier but it is still a very complex problem which had required me to put many iterations (so far it works nicely with my own statements)
For ticket parsing, AWS textract + parsing rules has been enough.
My integrated terminal + file browser with enhanced autocomplete and search. Replaces all native OS ones for me: https://github.com/David-OConnor/peregrine . Also includes misc utility OS functionality, tailored. (Advanced clipboard, a scratchpad etc)
Lately we've been working on access for AI agents. A lot of people want their own agent to do the repetitive part of time tracking. Control remains important, which is why we rebuilt our core this month to support change history and full undo/redo functionality. Feel free to try it with your own agent.
We built this after the timesheet tool we paid for raised its prices and moved some features that we used regularly behind more expensive plans. We had to add three new colleagues, and everything combined this would cost us $500 a year more for the same software.
We run Elixir and Phoenix LiveView on a Kubernetes cluster made of Talos instances.
We're still looking for feedback on the product, so please let us know what you think!
A private, paid, ad-free and personally customizable search engine: Uruky [1]. It was launched earlier this year and reached 400 MAUs and €2k MRR last month.
A simple EU static and app hosting service: PodBy [2]. It was launched last week, and is at €10 MRR.
- hermitator, a tool to run multiple isolated AI agents for WhatsApp on a single machine, with idle agents using no RAM: https://www.encona.com/posts/hermitator/
- showntell, which allows your coding agent to explain things to you with a video that starts playing within a few seconds of you asking: https://www.encona.com/posts/showntell/
- a philosophy tutor (testing in private for now) that does 2:1 live philosophy lessons, each exploring a single topic (aimed at ~10yo kids). I've tested this successfully (i.e. two kids did an hour and they want to do it again), and am trying to make improvements.
Oh, showntell is nice! I'll try incorporating it into my project (obeya.si), which already uses demo videos to show me work it did. But when I ask a question, the response is still just text. Could be so much better with your solution.
It combines nonograms with crosswords, requiring both logic and word knowledge to solve. I am still working on the interface and tutorial, as well as the puzzle difficulty levels.
been curious about the intersection between observability, SDKs, and product analytics. i couldn't find anything around SDK observability so built this little tool. curious what yall think!
i assume at bigger orgs there is some internal handrolled tool like this that ingests datadog/sentry/whatever info...if anyone plays around in this space and has feedback id love it!
I am currently on maternity leave with my first baby have just launched to the app store my first app!
I'm a product designer by trade and learnt to vibe code on Claude about a month ago now. I was weaning my baby and he developed a rash and haven't been able to work out what the pattern was causing it. So I created an app for weaning parents to log the food, add any reactions so when talking to their GP they can start to see a pattern and all advice, like nutrients, food variety, how to serve food all coming from the NHS and FSA.
Currently got 12 mums using the app and would love any feedback to make it even better! <3
For the past two years I've been working on my day planner app Ember.
Most planners treat your schedule as fixed, so the moment a meeting runs long or a task takes twice as long, the plan is wrong and you either fix it by hand or give up on it. In Ember, if you move a task or change how long it takes, everything after it shifts later, and it flows around fixed things like meetings. It's been super helpful and I have been adding new features ever since.
Right now I'm working on a native Kotlin app to replace the current one is web + Capacitor and exploring full duplex voice models so I can add or move things around the app while having a conversation with an agent.
I've got 40k+ photos on my iphone, and wanted a better way to interact with them. The basic idea I had was to virtually stitch all the photos together into a gigapixel image, and navigate that image with pinch and zoom. On import, the app hierarchically groups photos based on date and similarity. The user can go back and edit the groups, change the layout, and adjust the size of photos if they choose.
The interesting I noticed is that in the process of curating and organizing my photos, I stumble across memories and moments that I had completely forgot.
Development Challenge: The app does two interesting things underneath the hood... First, given an arbitrary number of thumbnails with different aspect ratios, it finds the best sized rectangle that completely fits all the photos. This allows me to recursively stitch groups together and always output a rectangular thumbnail for the group. Second, the pinch/zoom UI uses these nested thumbnails directly without slicing/dicing to fixed sized tiles.
1-year part-time experiment: can you build a large project with AI writing the code and AI plus deterministic tools verifying/guiding/enforcing it for structural and behavior consistency, without a human reading the code?
I'm building a polyglot web framework in 15 languages, with C# and Lean4 as the reference implementations. Structural and behavioral audits using deterministic tools check that every port stays consistent with the reference. It's about 1.8M lines now. I have not reviewed the code/neither have i verified the behavior tests except at the time of giving requirement and never seen some of the language code at all.
The question is how far can we produce "working code → correct code [Possible] → good code [Unknown]" with deterministic audits correcting structure, ai mistakes and behavior validation.
So far the quality looks acceptable with lots of correction of AI output done without looking, but it isn't fully verified.
A bit like iodine in the basic idea, but the execution and network topologies you build with it are different. It's more of a building block; you get your DNS channels represented as UDP endpoints, and take it from there, ie you can run Wireguard over it, or SSH or a SOCKS5 proxy (using quicmux, so over QUIC), or run Mosh straight over DNS if you're crafty.
It has an Android app too! I tried to submit it to F-Droid but they're so swamped in spammy or low-quality apps that they're not taking any brand new apps :-(
A sad state of affairs, that.
A small team and I are working on Sendery [1], a transactional email platform for product teams. We're trying to fix some of the headaches that come with managing product emails as companies scale: keeping track of hundreds of templates, maintaining consistent designs, coordinating changes between developers and marketing/ops, and managing everything across multiple languages is way more time consuming than it should be.
It's been a fun project to work on since it's been a pain point for myself on many projects, so I'm mostly just trying to solve my own set of problems and hoping other people find it useful too.
I've been working on a planner for the past 2 years for people whose day never goes the way they planned it.
I was trying to solve the problem I had with time-blocking, where your day ends up derailed and you give up on it because it takes too long to replan. I have solved with the idea of a cascading calendar. If I move a task down or change its duration it pushes other tasks later in the day.
Since then its kinda grown into my full productivity system for logging my thoughts and planning and documenting my life
Not really sure what to call it, Linux kernel ish implemented in JS so you can run it in the browser. Each program runs in WASM. Everything bootstrapped from a C/Cpp compiler compiled from source with a new target. Also kinda turned into Docker for WASM.
Recommended for people that feel a strong urge to run a browser inside a browser.
Also quite useful to run coding agents sandboxed inside the browser, with HTTP policies and "backend managed skills/extensions". Compiles Codex from source (has a rust compiler), runs pi agent and CC (the old node one) on a custom javascript runtime on top of quickjs compiled inside the environment.
Everything is static JS (or the robots are lying to me)
I’ve been working on Beacon, which just got Beacon approved on the App Store! It’s a mobile app to instantly answer “which friend is free to talk right now?” and summon them in a single tap.
Send your beacon to a group, and everyone gets a gentle ping. The first person who answers gets connected for a 1-1 call, and for everyone else the signal drops silently. It does not log any missed calls so there’s neither pressure to answer nor pressure to return calls.
Overall it works really well! It’s best if you're able to join with 4+ friends/family you’d like to chat with more. I have dozens of connections in the app now and it feels like absolute magic to me. I tap the button to send my beacon to everyone and get a random friend on the line.
I wrote a children’s story in Galician about open source. It's told as a fable. I made it open source, and people are now helping translate it into Spanish and English.
I’m looking for ways to fund printing and distribute the book. If anyone has done something similar, I’d love to learn from your experience!
For now Im working on an agent-orchestration application that is taking tickets from a tracker and giving them to an Agent. It's called Even Keel.
The main feature is, that I can use different subscriptions/API keys (Claude and Codex - v0.1.4) and trackers (GitHub, Linear and Huly - v0.1.4) to make Even Keel start an agent for every single issue, task by task, one at a time per folder. There is also a parallel feature, so I can run more than 1 project at a time.
Once upon a time, when codex agent was integrated into the app, I wrote a ticket to make codex his own hook. It reused a hook that blocked its own git merge, if tests did not pass. I started calling it "A new agent writes its own cage"... Right after, it turned out it doesn't listen to the hook and it fixed it
Version 0.1.4 is already shipped, but v0.2 is coming soon with more supported trackers, agents and new features.
How to make data storage on multiple storage devices unbreakable and consistent? The former means for me that back-ups should be automatical and serverless, the latter means that no matter how much of that HDD/SDD devices are connected, I want always to know all metadata of all files I have. I want this software under GPLv3 if to consider GPL still protects me from anything because permissive licenses are no better than WTFPL by any single case.
My first idea is to require the user to have more than one disk for the system needs and the system should know all of the rest data devices user owns. My second idea is to treat different files as having different priority which impose multiple copies. My third idea is to make the software accessible, and the last idea is to make encrypting of all the devices easy and with no ties to any commertial entities. I would like to have a discussion about how to not lose anything digital.
Magic AI - I'm using it as a sanitized kids chatgpt for 7 activities: Story, image creation, chat, reflections, journal, learn and quizzes. Sends me daily summary on Whatsapp.
It started with building pi in the cloud - with pure cloud flare workers. I have it working now, and can spin up - the wrangler agent can fully run it as long as you pass it an inference key, including a file view, a chatgpt like experience, and ability to build and host custom apps. Can use alarms to wake up to do plain language tasks, etc.
Trying to figure out how to package this up, and where this would even be useful for other apps.
Built an app to suggest best local beach for given weather for my town. My partner shared with local moms and they seem to be loving it - https://beach.metacursion.com/
Also revived old project - https://aminosplore.metacursion.com - the idea was to compare what essential amino acids each food contains using spider web chart. It’s been dead for a while due to USDA API changes so I revived, added macros and enabled to build foods into actual recipes.
Finally have been improving elsrc.com - an easy and scalable way to use ai image generators. If you need hundreds of images you can just request them by url. Supports versioning, batching, regional and national styling based on viewers IP and few other neat features!
I just got Beacon approved for the Apple App Store, it’s a mobile app for answering “which friend is free to talk right now?”
It’s a one-to-many phone call that only one person can answer. Send a beacon to a group, and everyone gets pinged at the same time. The first person who answers gets connected for a 1-1 call, and for everyone else the signal drops silently. It does not log any missed calls so there’s neither pressure to answer nor pressure to return calls.
Overall it works really well!
It works best if you're able to join with at 4+ friends/family you’d like to chat with more. I have dozens of connections in the app now and it feels like magic to me. I press the button and get a random friend!
I've been building out a link blog idea called Linked List that I've wanted for a long time. Curation will be even more important in the future and I would love to have another option beyond Hacker News and Lobsters to find good reads suggested by fellow humans. It has been a fun space to explore ideas.
Not quite ready to open up sign-ups, but you can get an idea of the site here: https://linkedl.ist/@kale
I am building a system that connects texts (books, research papers, etc.) in a meaningful way to help people decide what to read and be aware of how texts relate to each other.
The idea is that the source code is annotated with special rules which configure how parts of the code (also configs and docs) depend on each other. This can be done either directly in the source code comments or in a config file blockwatch.toml
This tool is very useful for large codebases that e.g. share constants across programming languages and configs, have docs that should strictly reflect the current state of the code, etc
It uses tree-sitter to parse the source code, 35+ languages are supported.
The project is actively developed, your feedback is welcomed! Thanks!
I am building Abbox trying to help engineers focus longer on tasks without being interrupted by non-technical people with small requests.
I always got annoyed by change this h1 to h2, add tracking to share button, why can’t the customer play the video, the styles don’t match figma.
If you are engineer who also gets distracted I would love to hear your opinion and feedback. It is a bit hard to put it all together as every department has his own needs and opinion about the website.
Thesoapbox.email - a social experiment email lottery to see what people do when suddenly given enormous reach. Every day a new subscriber is drawn to email the whole list. It’s hit 10,000 views in a couple of weeks which is my most viral project so far.
The second - a personal AI learning platform built on recursive self improvement and tailored content to push my boundaries. I’m a lifelong learner and this unifies everything for me.
(Currently not public)
I've been working on an Obsidian plugin to quickly enter caloric and nutritional data (https://www.github.com/Krishna-Sivakumar/joule). The core features are done (A DSL and a runtime for finding and resolving formulae recursively), and I need to work on documentation and some convenience features. Hoping to finish it this month!
Maintaining my indie social network for tech people https://stackness.dev/ (currently struggling to keep users on the platform, even games and achievements didn't help, can you believe that? :D)
and this weekend I'll finally launch a tool for guitarists and their friends https://kantamigo.com/
I've been working on a voxel game engine called Bonsai for ~10 years.
Probably the most interesting thing about it at the moment is the editor. The world, and most things in it, are represented as collections of SDFs. More accurately, they're density fields, but, potato-tomato. One neat thing about representing the world as SDFs is they've been thoroughly researched and documented by a guy named Inigo Quilez, and they have a lot of nice mathematical properties. For example, you can do a smooth union of arbitrary SDFs to get nice rounded contours where shapes join.
The world generator and editor operate almost entirely on the GPU. The terrain and editor pipeline is a three stage process. The first stage is user-configurable; users write an arbitrary 'shaping' shader in GLSL. The second stage is a fused stage; it does a convolution on the output of the first stage to compute a normal, runs a second user-configurable 'decoration' function, and finally applies edits. Edits are collections of SDFs that are 'compiled' from a graph structure on the CPU into a bytecode, and interpreted in the shader. The last stage runs another convolution to compute the final normal, then packs an occupancy bit, color, and normal down into a 32-bit representation, suitable for transfer back to the CPU. I expect I will expand this to include additional user-level bits for encoding biome type and GPU-driven spawning.
The entire shader pipeline is one fused 'ubershader' .. everything is bound at startup and the shader is dispatched three times, incrementing a 'Stage' uniform on each dispatch. This strategy conceptually makes the driver overhead nearly zero, and the profile agrees.
The final step in the process is giving the CPU a chance to spawn additional entities and structures after generation. This is not completely fleshed out, but the bones are there. The idea is that the results of the GPU work are passed back to the CPU, which can run arbitrary callbacks to spawn additional edits and entities. Once spawning is complete, the result is meshed using a fast binary meshing algorithm and uploaded to the GPU. I keep 16 bits of data per voxel from the GPU work, 1 bit of occupancy, and 15 bits of normal for collision response, which works out to half a megabyte per chunk. At the moment I only keep chunks which contain a piece of the world surface, and discard the rest. Surface chunks are dense, and may be compressed in the future, but keeping this amount of data scales to world sizes greater than 1B voxels cubed, which seems good enough for now.
In terms of raw performance, GPUs are ridiculous. My 3080ti mobile gets somewhere on the order of 500B-1T perlin samples per second with a naive perlin implementation. Compared to FastNoise2, that's about 1000x more, which sounds wrong since my GPU only has 35x more capacity, but, that's what my initial profiling said. That number is the marketing number; the real throughput is significantly less due to multi-stage nature of the system, but, there's a lot of headroom for improvement. For example, dropping the backing textures from 32bit to 16bit roughly quadruples performance. Since we only really care about the surface where values are near zero, I believe using 8 or 16 bit is a viable option, though this claim will require a fair amount of study to validate.
All in all, I'm pleased with how this is all working out. My original goal was to write a highly flexible system giving users both easy-to-use tools and granular control over the final output. I think I've done a fairly good job at that, and the fact it's also super fast is the cherry on top.
I'll do some shameless self-promotion and leave some links here for anyone interested in looking at the engine, language code, or some pretty pictures.
That's so cool, building something consistently for 10 years is no small feat. How did AI helped you, or How did AI impacted your day to day work in recent years ?
Thank you :) One of the nice things about working on a game engine is it's actually a large collection of small projects disguised a big one. That let's my squirrel-brain jump from thing to thing and have enough variety to not get bored.
I just started using AI a few months ago. My smoke-test was "can it write a simple memory allocator". Until recently, every time I tried to get an AI to write an arena or simple buddy-style heap allocator, they just vomited all over themselves. To my surprise, the last time I tried, the AI absolutely crushed it; it unearthed tricks I didn't even know about in one shot. Since then, I've been using AIs to do the boring, rote work that effectively just boils down to me typing in something that's in my head. For that, they're great. I typically sketch out the thing I want it to make, then tell it to fill in the blanks. Sometimes they still go off the rails, but all in all it's a useful tool.
Where I've found they still struggle is making any architectural decisions, or where performance matters.. even a little bit. They often make super braindead decisions when it comes to performance for no discernable reason. They'll course correct when you tell them to, but I've had to keep a close eye on the output and reign it in frequently. Asking an AI to design a system from the API level hasn't ever returned a good result for me.. I always do that work.
All in all, I feel like using AI will probably make me a better programmer in the long run, contrary to popular belief. They accelerate the boring stuff I already know how to do, which leaves me with more time to work on the edge of my skill level where I grow as an engineer. Guess we'll see in 5 years.
Hence (https://hencegame.com) is a modern twist on Einstein's puzzle, with fun picture clues instead of text. It's a free online daily puzzle, the same for everyone, where every puzzle is always solvable by pure deduction with no guessing.
I launched it October 5. The hardest part was the graphics, making each picture clue unique enough that you recognize it immediately and know what category it's from, even at small sizes. Easier said than done.
Every puzzle is run through a verifier first to make sure it is solvable without any guessing, just by pure logic, so there is always a next step to find, which is what makes it feel so rewarding. There's a hint system to help and it works really well if I do say so myself . No account needed to play.
From what I've been hearing so far, the people who have checked it out are enjoying it and can't wait for tomorrow's puzzle, so that's a good sign.
I'd love to hear how the difficulty feels, and whether the pictures read clearly.
I've been working on a small search engine for content creators that play and/or review indie games. It's written in C and has been a lot of fun to work on.
I keep saving blogs and articles but almost never revisit them.
So I’m building an Android app designed to help me actually read what I save. You only get one saved article per day. Finish today’s read to unlock the next one. You can also save Twitter, Reddit, etc. posts separately and revisit them anytime.
Right now I’m mainly validating whether people would actually want something like this. If there’s enough interest, I’ll run a small closed beta for Android.
I've been working on trying to be more helpful around the house, and trying to spend more time with my kids. I'm trying to give my partner a little bit of relief so that they can pursue their passions as well.
So far it's helpful for messages between my own AI's (chatgpt and claude etc), and for having my AI trade tech tips with a friend's AI or talk to my wife's AI. My AI also got its first message from a strangers AI from france which has also been really interesting. I read all messages first and approve before sending anything, but otherwise they're just talking about whatever they want.
I'm working on Hale, an abstract distributed systems language.
It's a traditional systems language in the sense that you have fine control over how resources are used, but it's abstract in the sense that you have to come to an agreement with the compiler about your intent around the resources, rather than directly declaring the machinery.
It's a distributed language in the sense that there is a higher level structural abstraction in the language that situates computations of the program within a recursive graph structure. Part of that structure incorporates message broker semantics. This gives the compiler a unified view of programs across arbitrary process boundaries. The structural abstractions also enable an effects+formal verification syntax within the language, which allows programmers to codify architectural decisions that are then checked by the compiler.
The compiler itself is infrastructure aware, and the highest layer of functionality it handles is configuring the environment into which applications deploy. The single language handles everything from individual memory allocations to distributed fleet deployments.
The language is close to v1, and I'm currently building a fully distributed inference-as-a-service application with it. Arbitrary hardware/supply matched against arbitrary callers/demand in real time.
I’m building Tinta, a journal app for Android where all the AI runs on the phone. It works in airplane mode once the models are downloaded.
You write or dictate an entry, and a small model (LFM2.5 1.2B) asks you a question to reflect further. It turns out 1.2B models are decent at asking questions, so I leaned into that. An embedding model finds older entries about the same thing and shows bits of them under the new one.
Most people using it are on budget phones running the models on the CPU, where the app provides good experience.
Great idea! I'd love this, but as an open source app. Out of interest (not a criticism), what made you decide to do this closed source, especially considering there doesn't seem to be a commercial interest involved (for now)?
Thanks for a good question!
I see this app as an MVP that needs user feedback to deliver exceptional value in a world where cloud AI is abundant and doesn't use 2GB+ of RAM to run. I am excited to see what shape it will take, and it feels too early to determine source availability.
GeoPeek, a small macOS Quick Look extension for GeoJSON. I work with geospatial data and wanted to press Space on a .geojson in Finder and see a map instead of raw JSON. It also shows an attribute table and basic stats. https://github.com/serifsadi/geopeek
Still working away on my book (and library book) recommendations service https://leafle.org
It has if nothing else been very useful for me (it has correctly deduced that I would get something out of borrowing Grief is the Thing With Feathers, the Dispossessed, and Cosmicomics, none of which I would have thought of reading). Plus it has been a fun opportunity to play around with LLMs as an interface and a research and reverse engineering tool; as well as Jev for non-user facing quality work like banning crap openrouter providers based on their output before putting it in front of a user.
Most recently I’ve been dropping off flyers for it at a few local libraries in my area.
I'm working on a TradingView Pine Script compatible Python framework: https://pynecore.org . And a Pine Script compiler for it + a full IDE for Pine Script with syntax highlight, language server, full featured debugger, charting, data downloads: https://pynesys.io .
It is 99.771% bit exact on the top 1000 Pine Scripts and 100% trade match.
I'm doing it solo.
* files in organised trees/graphs generated using LSPs and symbol references instead of linear file list
* keyboard shortcuts for everything from navigation, pan, zoom, comment, note, submit...
* mostly github compatible with native features
* LSP powered symbol resolution for inspecting definitions and references
* LLM powered reviews supported
* Inline LLM question answering
We just got a puppy, and needed a way to track potties to help us with house training. So, with AI/Claude's help, I found a similar open source package (Baby Buddy), forked it, and very quickly made it my own. Self-hosted, local for privacy reasons.
An Obsidian plugin that turns handwritten reMarkable notes into searchable text. You can run the handwriting recognition locally on a Mac. The latest feature I added sends your highlights and margin notes from the tablet to Readwise. https://taggedsync.com
I built the original for myself because I wanted to listen to acoup and other history blogs while sleeping. And in the spirit of getting better at finishing projects decided make it public even tho it’s nothing really unique or special.
Still lots to polish especially in the text/copy and ui but the core features should be working.
Working on a natural language intelligence layer for contacts - "Pull everyone I know who works at a Series A healthtech in the Bay Area who also went to Stanford for ungrad" => either structured or unstructured results. Large-scale data support, automatically enriched with up-to-date data, fast/accurate results. Integrates natively with Gmail and Gcal.
Looking to pilot for free with small qualified teams (early-stage startups with real/immediate need).
It uses Apple Intelligence to help brainstorm domain name ideas. I’m interested in seeing what small local models can do as they catch up to larger ones, this is my first iOS app so I started with something simple that scratches my own itch.
Building Wishfox [1] for around 6 years. It's a wishlist app with price tracking, groups, secret Santa and more. Recently did a major rebranding, and the mobile app is almost ready for testers.
Then I'm working on improving my two open-source apps, Portfox [2] and Localfox [3]. Portfox shows ports for running services and dev servers, Localfox provides them with .localhost domains with TLS and Cloudflare tunnels.
Currently we are working on Kumoss Open Source, intent-based infrastructure platform that turns natural-language requests into infrastructure changes, while keeping validation, compliance and existing engineering workflows in the loop.
Basically allow developers to request any infrastructure need in a web portal and automatically that request is fulfilled following the organization standards the platform team defined.
I'm vibing an old style RTS game and exploring how far can I get with local models for drawing, 3D models, rigging and sprites. I love how much I can accomplish with Claude in terms of code, but gluing it all together with graphics still requires plenty of "manual" work.
Werkbook (https://werkbook.app), digital forms and photo reports for field crews. I'm a solo Rails dev, and thanks to Claude it now has native iOS and Android apps instead of Hotwire Native web-view wrappers.
I've just released v1.0 of Rouler which is VPN split tunneling for windows. Think IP rules for Linux, but with per-app/service connection steering. The whole app is built in Rust, including the kernel driver. Having released v1 I'm now starting work on adding folder and windows store support. https://rouler.io
I’m still working on https://leafle.org which is a tool for getting more interesting reading (or listening) suggestions that one can borrow from the library (or Libby). I’m done for the moment with building it, so I’ve been doing things like making flyers and leaving them in local libraries.
Been a fun opportunity to play around with ai in various ways - mixed but definitely powerful experience in using it for data research (finding libraries and their locations), very impressive on reverse engineering apis to get book availability. And then a lot of interesting stuff around getting good behaviour from LLMs as chat partners, for example using jev to improve results from using openrouter by getting it to check for common failure modes and ban providers based on their output.
For the month of November (and starting in October this year) I like to hold weekly cooking challenges to practice and refine my Thanksgiving day feast. Now about 3 weeks out from the final boss level of cooking, and I'm having a week all about brussel sprouts. Earlier weeks were cauliflower and the classic seasoning blend of sage, rosemary and thyme.
I am building Cloud based context compiler and intelligent codebase visualizer for Developers and AI Agents, for easy PR review, onboarding and easy architectural understanding
My main project is trying not to lose my mind, given all the things I can now create thanks to AI.
Jokes aside (although I'm not entirely joking about my mental health), my main project is https://auradistill.com: First-principles thinking for your AI agents.
It's not just about giving your personal AI agents "memory". It's about encoding memories in a way that can handle ambiguity and contradictions.
That project alone has got me completely hooked on building for and with LLMs.
I've got other projects too, but this is probably the one I've poured the most time and energy into.
https://hatamoto.dev - a complete replacement to Wordpress and its most used plugins built into the system. One go binary, no node , no compilation, Postgres and sandboxes plugins. Built in AI analytics
I'm working on a platform for selling event tickets (kind of an alternative to Ticketmaster or weezevent).
I'm working on it on my free time for now (I am also using it to test working with AI tools ).
Not sure yet if I want to build a company or not.
Aficionado (https://aficiona.do) - a place to discover, scrapbook and remember media (movies, video games, tv shows, music and books) that resonated with you or that you'd like to try.
working on AudioWeave, an ad free, offline first Android podcast player.
switching from iOS, I wasn't happy with the existing podcast apps, so I built this mostly for myself. I've been using it for the last six months and finally pushed a public beta.
My day to day is BE so I used the project to learn Kotlin and Jetpack Compose. Everything is strictly native using Room and Media3/ExoPlayer. I wanted a nice looking (subjective), privacy-focused app without any tracking or ads.
I'm building a self-hosted media server for video, like Plex or Jellyfin. I'll be opening the beta in about a month.
Features:
- No paywalls, bandwidth limiting, or crippled behavior. No login required.
- Single go binary, deployed via Docker.
- Reaches your home server from anywhere with no port forwarding, no VPN and no paywall, even behind carrier-grade NAT. No Tailscale required.
- Cast to another device. You can use any logged in device as a playback screen.
- Watch together, with voice support: Play synced video with friends, and also chat voice. You can also mix this with casting to another device.
- Dialogue boost and loudness levelling for every film.
- Real adaptive quality, tuned for a home's upload.
- Hardware transcoding, full quality in every supported browser.
- A library that explains itself: what the scanner saw, why each title matched, guesses marked for you to check, and a disk that's gone missing shown as missing instead of its titles silently vanishing.
- Your data stays yours: The database is a rebuildable index, metadata corrections can be written back as NFO files that Kodi and Jellyfin read, and an automatic backup keeps what can't be re-derived (watch state, preferences, shares).
Security is a primary concern:
With the ever increasing pace of hacks and CVEs and AI-assisted intrusion, you simply can't be too careful anymore.
- The core (which talks to the network, handles the API, users and the database) never opens a media file or runs FFmpeg.
- A separate media node does every file read and every transcode. It has no network access and talks to the core only over Unix sockets.
- FFmpeg runs under Landlock and seccomp with no path to the file and no socket. All it can do is read and write streams.
- Small dependency tree on an allowlist, pure Go with no cgo, secrets sealed at rest and never logged.
- The library can be mounted read-only, so even a full compromise can't encrypt or delete your media.
- Cryptographically signed chain for servers and releases.
But apart from that, I just updated my website's theme to something I wanted to do for quite a while now – a fantasy desktop!
If you'd like to check it out: https://0xff.nu
I'm also trying to just learn new things with the time I was refunded by being unemployed.
I’m based in Berlin and currently working on ChatFrom.io . You upload your knowledge base, and it uses advanced RAG with hybrid search and reranking to find answers when someone asks a question. It connects to Stripe, Shopify, and other tools, and works on Telegram, Slack, and more. Free to start.
working on talk deep https://talkdeep.me, a little app for keeping in touch with friends after i moved countries. you pair with a friend and get one question a day, or go through decks of deeper questions together. i've never done an app before so building something from scratch has been a really fun creative outlet
been building BuildOpsy mostly writing about system design, infrastructure failures and the occasional “well, that was an expensive outage” postmortem.
Also building calculators for things engineers usually calculate in a spreadsheet at 2am. you can see some on the site(https://buildopsy.com/calculators.html)
There is already a site here where you say what your job is and garages submit their offers. They can also auto-submit offers if it’s something routine like a standard service.
For years, I struggled to make any headway with Linux - I get so frustrated at how obscure, inscrutable, and unintuitive TUIs are! God I don’t want to waste my life typing esoteric strings of commands into the terminal. When I started with computers that was all there was, and once GUIs happened I never looked back.
Good news though, these days Claude will just do it for me!
I’ve been working on setting up a home media server, using headless Debian on a pi - I’ve got it set up to stream music and videos, hold all my ROMs, I’ve got it automatically standardizing file names and handling tagging and subtitles - all attached to a big 6-drive enclosure that I just stuffed all my random old HDDs into.
Even though I haven’t written more than a dozen lines of code or commands for the whole thing, I am pretty fuckin pleased that now I can just wave a magic wand and say “I want all my video game soundtracks filed under the game name rather than the artist” and Claude uh finds a way to grant my wish. I spend about an hour each night tweaking it and it’s so easy to customize to be exactly what I need.
My husband created an AI-powered website, and I've been using it to generate engaging AI videos for social media. This not only helps promote his website (seedeo.ai) but also expands our website's content library, which gives me a great sense of accomplishment.
I'm nearing the end of a somewhat long and arduous journey of writing a GPU-accelerated voxel terrain generator and editor.
The terrain and editor pipeline is a three stage process. The first stage is user-configurable; users write an arbitrary 'shaping' shader in GLSL. The second stage is a fused stage; it does a convolution on the output of the first stage to compute a normal, runs a second user-configurable 'decoration' function, and finally applies edits. Edits are collections of SDFs that are 'compiled' from a graph structure on the CPU into a bytecode, and interpreted in the shader. The last stage runs another convolution to compute the final normal, then packs an occupancy bit, color, and normal down into a 32-bit representation, suitable for transfer back to the CPU. I expect I will expand this to include additional user-level bits for encoding biome type and GPU-driven spawning.
The entire shader pipeline is one fused 'ubershader' .. everything is bound at startup and the shader is dispatched three times, incrementing a 'Stage' uniform on each dispatch. This strategy conceptually makes the driver overhead nearly zero, and the profile agrees.
The final step in the process is giving the CPU a chance to spawn additional entities and structures after generation. This is not completely fleshed out, but the bones are there. The idea is that the results of the GPU work are passed back to the CPU, which can run arbitrary callbacks to spawn additional edits and entities. Once spawning is complete, the result is meshed using a fast binary meshing algorithm and uploaded to the GPU. I keep 16 bits of data per voxel from the GPU work, 1 bit of occupancy, and 15 bits of normal for collision response, which works out to half a megabyte per chunk. At the moment I only keep chunks which contain a piece of the world surface, and discard the rest. Surface chunks are dense, and may be compressed in the future, but keeping this amount of data scales to world sizes greater than 1B voxels cubed, which seems good enough for now.
In terms of raw performance, GPUs are ridiculous. My 3080ti mobile gets somewhere on the order of 500B-1T perlin samples per second with a naive perlin implementation. Compared to FastNoise2, that's about 1000x more, which sounds wrong since my GPU only has 35x more capacity, but, that's what my initial profiling said. That number is the marketing number; the real throughput is significantly less due to multi-stage nature of the system, but, there's a lot of headroom for improvement. For example, dropping the backing textures from 32bit to 16bit roughly quadruples performance. Since we only really care about the surface where values are near zero, I believe using 8 or 16 bit is a viable option, though this claim will require a fair amount of study to validate.
All in all, I'm pleased with how this is all working out. My original goal was to write a highly flexible system giving users both easy-to-use tools and granular control over the final output. I think I've done a fairly good job at that, and the fact it's also super fast is the cherry on top.
I am working on a Rust-based ext4 driver. The idea is that it can be re-used as a FUSE-based driver as well as (Windows) kernel driver (eventually).
Most of the essential features are implemented so far, but some (mandatory) features are lacking (xattrs support). My goal is to get to a full ext4 implementation that also supports ext2 and ext3 read/write.
The initial motivation was to have a pet project with a use-case for Rust, as I wanted to improve my Rust coding skills, after I did some work on a game server implementation in Rust earlier [1].
My goal is to have an implementation where I have digested every line mentally, which is something that pretty much excludes slop-like use of AI for large scale code generation. And of course I want to really trust the code that touches my files via raw hard disk access.
Hence, I use AI mainly for unit tests and (scoped) code rewriting when my code gets too verbose and cumbersome.
In the long run I'd want to add fuzzing to the implementation, so if anyone has experience on this, I'd be interested in getting some pointers.
Currently working on adding support for iPhone Duo to my hacker news app Hack. The app has been around for 9 years and when I made it, the menu bars and navigation buttons were custom views. That doesn't seem to work well for iPhone duo due to its vertical stack menu on the right side. So, I've been switching over to iOS native navigation bars and menu bars along with a bunch of other improvements. I was hoping to have it done couple weeks ago but I had to leave country for a family emergency. So I am working on it this week. It's mostly done, I'm just doing some final testing now and will submit to Apple this week:
Ur is an (in-progress) universal formatter and linter for every programming language. It's in preview right now, which just means friends are messing around with it and having a good time pointing out all the bugs. Docs are mostly up to date, but buyer beware.
Edit: ok, that's done and I did my best to normalize the timestamps for the posts from the older threads, but it's a bit messy. If you see weird things that's why.
I'm so fucking tired. I built this tool, it's great, I love it, but its full capabilities are too complex to udnerstand. Because of that I struggle with marketing, as everyone is too lazy to try it out. Because of that I get no feedback, nothing to improve, and I can't make the tool any simpler, as otherwise it defeats the magic behind it.
Software is complicated by nature. It's our job to make it easy for the user.
Take a look at FFmpeg as an example. It's a complex tool that can do most things audio/video. A power user can do magical things, but they still provide simple commands for the most basic use-cases like conversion or trimming.
I've been working on online payment rails for businesses, developers and agents called Tuppence (https://tuppence.ai).
Accept online card payments from just 0.79% (UK) and earn 0.89% on all card spend [1], or get paid out to your bank.
I'm building tuppence to be available in the US, and 20 EU countries.
I'm getting a Point of Sale (POS) manufactured in the UK [2] for those who want to accept payments in person with card readers.
My goal is to help people sell their goods and make a living without paying over the odds for the privilege.
I charge a competitive fee for payment processing, give money back on card spend (with no cashback caps/limits) and do not charge saas fees or lock people into contracts.
Tuppences currently allows people to sell online with a store builder, expense management, payment links, direct debits & bookings - essentially Tuppence is a 'business' layer on top of what you get from the likes of Stripe, Adyen and Checkout.com. There's also have a 'connect' platform, so you can roll this out to your marketplace business.
Next on my roadmap is ticketing and 'treasury'. Treasury i'm personally excited about as you'll be able to earn 4% yield on your earnings, FDIC Insurance (FSCS in the UK), multi-currency banking in fiat balances such as USD, EUR, and GBP, or stablecoin balances like OUSD and USDC, and have bank account numbers to send or receive funds.
---
It's brand new, i'm still testing payment -> payout flows. These take a little time to verify on live payment rails.
Happy to give some fee-free payment processing for those who reach out and want to be early adopters, I can be reached at ollie@tuppence.ai, or comment on thread.
---
Tech Stack TLDR:
* TypeScript monorepo covering the API, dashboard, docs, mobile app and SDKs.
* The API runs on Ts.ED over Express with TypeORM and Postgres.
* Background jobs use pg-boss, so the queue lives in Postgres and there’s no Redis or separate broker.
* The dashboard, till and storefront are React + Vite + Tailwind, the docs and help centre are Astro, and the mobile app is React Native/Expo. We ship SDKs for Node, Python, iOS, Android, React Native and WordPress, plus an MCP server and agent tools.
* Everything runs on Google Cloud Run and Cloud SQL.
Guaranteed weight loss within 2–3 weeks; many people start seeing results within the very first week. 17,000+ members already!
Guaranteed results are achieved using the state of the art adaptive TDEE (total calorie burn of the day) tracking technique we've developed. It's sensorless, so no watch or other wearable is needed. It's way more accurate than those wearables (data included on the website), and our group has 2,000+ members ( https://macrocodex.app/community/ ). You can also read Play Store reviews.
MacroCodex also works on both Android and iOS and any platform where PWA app can run.
Other other 2 apps are Android only (for now):
If you are looking for a workout app with Alpha Progression or MacroFactor Workout-like “Smart Progression,” or an app where you can run any program like Boostcamp but ad- and subscription-free, check out the Symbiote workout app (beta)
https://symbiote-studio.macrocodex.app/
IF you are looking for AI photo Calorie tracking with lots of features which make it very fast to log food on daily basis checkout CalorieCodex (beta):
https://caloriecodex.macrocodex.app/
All the apps have been under active development for a few months now.
I’m still looking for work in the mean time but I’m proud so far of what I’ve achieved.
https://getcrewbird.com/
making it feel fun versus a basic quiz workflow is the hardest part.
https://salesfighter.app
A singing training app for bad singers. You sing the pitch line and get a score. Let's you sing to any original song by listening along and not playing the song itself.
Recently got some good feedback on HN https://news.ycombinator.com/item?id=50195488 and am following up with Linux and Windows versions and using Fable 5.1 to internationalise the app and landing page which explains the app.
I finished the basics a long while ago. Since then I've added an AI agent, a public booking page, and support for multiple staff. Hopefully, I'm launching soon. Currently finishing final touches, gathering feedback and trying to figure out how to market this
University of California San Diego has deployed our Wayfinder and Waystation web apps for their 35,000+ students!
We're getting ready to launch our Vehicle Positions software, which replaces expensive, proprietary vehicle location systems with free, open source software that runs on commodity smartphones.
Our next-generation server software, Maglev, is adding support for newer transit data formats, like GTFS-Flex so that we can show demand responsive transit systems on the map.
We are always looking for more developers, biz dev, ux designers, product managers, and more to help out!
Open volunteer positions: https://ossvolunteers.com/organizations/open-transit-softwar...
Our software: https://github.com/OneBusAway/
More about us: https://opentransitsoftwarefoundation.org
Now we don't deploy this to private companies except some gov projects. But we are engaging existing IaaS and dedicated servers providers. Later hope we would fight with Broadcom and IBM :)
We previously ran a services company that built over 100 private clouds, then decided to turn that experience into a product with an aggressive level of automation to let DevOps, SRE, and platform teams hit a 1:300 ratio per developer (now dev + AI agent teams).
We are currently deployed across 12 datacenters and raising our seed round this month.
Regarding been curious, today I have created my first explainable video with Remotion +claude so if you want to see open edit results check https://promo-review-shalb-promo.stage.kube-dc.com/
Would be glad if you comment. Thanks!
Send your beacon to a group, and everyone gets a gentle ping. The first person who answers gets connected for a 1-1 call, and for everyone else the signal drops silently. It does not log any missed calls so there’s neither pressure to answer nor pressure to return calls.
Overall it works really well! It’s best if you're able to join with 4+ friends/family you’d like to chat with more. I have dozens of connections in the app now and it feels like absolute magic to me. I tap the button to send my beacon to everyone and get a random friend on the line.
https://trybeacon.chat/
https://testflight.apple.com/join/mqfhpmHY
The original goal was an alternative to ChessBase, but with an Apple-native GUI and iCloud sync between devices, and as an improved alternative to tools like chessbook.com for operating repertoire management and spaced repetition opening memorisation. In recent months I have focused more on the chess-playing parts of the app, trying to implement human-like computer opponents at all levels, using a transformer-based model (similar to the one described in the Maia 3 paper) combined with a very small search.
Recently I’ve also been working on chess variants. Since a while, the companion Apple Watch app plays 5x5 chess. More recently, I added a 10x10 chess as an Easter egg game inside the app, and I’m now working on hexagonal chess. With the current state of the art in computer chess, it’s almost trivial to implement vastly superhuman strength in all these variants, but implementing weaker levels that are enjoyable opponents is interesting and challenging. Unlike in normal chess, there are no big databases of human games to train on.
Also, creating interactive ways to view the experiment and visualizating the results.
https://crux.lakin.dev/eval
Project #1: I've been working on migrating an application [1] from GCP (where I pay $200/mo to host it) to a physical server (PowerEdge R7625) I've racked in a colo near me, running on my own IPs/ASN [2] and all that jazz. I've been dragging my feet on the this as there's just a lot of moving pieces, but I got a good chunk of the last few hard bits done today.
Project #2: Building a DIY bright-ass lamp, like Brighter [3] or Rhythm [4], but for a few hundred bucks instead of $1-2k. I've specced out the parts, designed the hardware, and built the firmware. Later today I'll do my first LED test against a bench supply, to get some hard numbers for if my cooling solution (a chunky aluminum heatsink and some Noctua PC fans) is sufficient for the full 500+ W final version.
As far as curiosities go, I've been doing a lot of Go (the board game, not the language) puzzles, and reading about sailing (mostly the book from the Hundredrabbits folk [5])
[1] https://abound.art
[2] https://bgp.tools/as/55078
[3] https://getbrighter.com/
[4] https://www.betterwithrhythm.com/
[5] https://100r.co/site/busy_doing_nothing.html
https://santawho.com A funny holiday game you play in person with the whole family on just one phone (the parent interviews everyone for jobs on the North Pole and surfaces inner insights about what everyone might enjoy from Santa). It’s for any age, private, ads free, just pure holiday spirit.
https://www.dreamlist.com - collaborative shopping lists and online wishlists and gift drive platform. Again ads-free, private, stack multiple lists into groups of lists to share on social media or with grandparents.
A number of other ones about to launch as well. All little tech, not big tech - private by default, no ads, no feeds or media, just surfacing what is needed.
https://github.com/jnz/inslib
Here's a demo: https://non.io/video/linen-demo.mp4
As a disclaimer I cut 32 seconds of generation time out of that video, but still I'm quite happy with the speed. Not launched yet, but it's been fun to work on / wanted to show off progress :)
The theme is an empty Earth. All natural Earth with no human constructions (until the players start playing and occupying it).
Using Rust and RocksDB.
eremoterra.com
Interactive breathing exercises. Working towards a medical device version for asthma and COPD.
I make a special `textual DSL` where I model business scenarios like `Make a sale that include 3 invoices : 2 to different cities in country A in currency USD and other in country B in currency EUR` + all the intermediate steps like transportation, manufacturing, etc.
I`m now in the woods with a simplification for the price engine, I dislike much how complicated is normally, and I think I have good enough solution for most cases that generate without dynamic calculations.
Product: webfrolic.com
Demo: demo.webfrolic.com
What it is?
Consider Okta (Workforce identity)+ Auth0 (Customer Identity), Authzed (authorization) + ServiceNow (IT service Desk) + Zendesk (Customer Support Desk) architected together
Being an engineer there is this inherent habit of prioritizing correctness over marketing so I would not oversell it, though its probably the most secure and complete thing that could be built from identity perspective
Status (WIP):
1. I am still debating/settling on how to incorporate the AI Agents needs into the architecture, big focus on avoiding the creation of security or performance foot guns where administrators leave room open for abuse with configuration despite all this attempt to build secure system
2. Feature stress testing and performance testing
This one I (mostly) did without LLMs, but the other project is vibe coded, but I'm enjoying it. An RTS made by Claude in Unity: https://github.com/SiGe11/StarForge2
Each game tick is an hour and games can last over a week. I got hooked on both Subterfuge and Neptunes Pride and always wished there were some more titles in this genre.
Currently in closed alpha.
https://ashdomes.com/
SuperBlocker is a native blocker that works on iPhone and iPad using Apple's latest URL filtering technology that blocks URLs without revealing it to Apple's relays or SuperBlocker's servers.
JSON Query: https://jsonquery.app (https://apps.apple.com/us/app/json-query/id953006734)
The original JSON Query app on the Mac App Store. Store, organize and query all of your JSON with jq or natural language. All of your data stays local so sensitive JSON can stay on your machine.
- removes the stuff that I don't care about (Firefox sync, VPN, etc.)
- fixes a lot of the awful UX/UI choices that the (frankly) incompetent product team at Mozilla refuse to
- actually adds stuff that I do like integrated mail and calendar, a document library system (basically Notion), native ad-blocking, terminal, task-tracking like Trello, time-tracking like Harvest, etc.
- has integrated AI (Apple intelligence for MacOS for the easy stuff and Claude for the hard stuff
It means that instead of having half a dozen applications open, I can just work within my Firefox fork. Because it's all deeply integrated, I can do wild stuff like:
- tell my AI to search the web on a schedule and send me a custom daily brief with my own email
- maintain a comprehensive memory of all of the work that I do, across half a dozen clients, and be able to query it easily
- pull analytics from multiple web dashboards and format them into something cohesive
- program the web with a node-based infinite canvas that has live websites as part of the workflow
- manage my tasks across multiple domains (for example, it will see that I sent an email to a client and automatically update a task that says that I need to do that)
It's incredibly powerful and it has made me much, much more efficient.
https://limittox.github.io/south-park-animator/
It’s open source: https://github.com/nicobrenner/id-le
I'm Canadian but I moved to USA to work for Microsoft. One thing I liked about the USA was how easy it was to order bloodwork/lab testing. My original idea was to bring that experience to Canada, but since starting I've found other ways to improve the patient healthcare experience in Canada too.
My company is https://teletest.ca
Still not sure where this is going as the space is super crowded and moving fast, but it's solving a lot of my own problems and doesn't rely on third parties to run.
[1] https://colliepwa.dev
I needed a code-first approach for making installations and interactive art. MaxMsp gets messy with very big projects. Python didn't really work well for music making and Tidal/Pi/Sclang are only meant for sound. So I needed something that has bpm synchronized threads, but can draw and do anything else like file IO, networking, hardware control etc. It works quite well now.
What I need is something to switch the account where the usage is taken from, not the configuration.
I made something simple & transparent named Switcheroo: https://github.com/FTCHD/switcheroo
Works with other AI CLIs like Gemini, and other various CLIs like Vercel (if you're managing multiple accounts like me), Netlify, Github etc
"Yes, OpenAI explicitly supports having multiple paid ChatGPT accounts. Each account has its own subscription, conversations, and usage allowance."
When playing Zwift, Rouvy, etc, I find that I get too competitive and ruin longer workouts by trying to keep up with or sprint past people who are much faster than me over long distances.
This is bad for building stamina and endurance, but I get bored, and it's hard to fight boredom and physical strain both at once.
I do not get bored sitting in front of Balatro, Raccoin, Tetris - anything with bright colours and satisfying noises and combos.
The first finished prototype game is very simple: you sort coloured marbles. Higher watt output from the bike means marbles are blown, by wind, upwards as they roll across the screen. Lower watt output means they fall. Steady watt output means they move horizontally. Marble goes into correct hole -> satisfying noise. It rewards steady watt output over time, with minor adjustments, which is what I'm trying to work on right now.
I was pretty amazed by how engaging I found it (total iPad baby brain). The miles really fly by - I get tired before I get bored.
I think that next I'll either do a tower defence game, or a less-dopamine-rewarding game in the style of Kerbal Space Program. Taking your actual weight and watt output, putting a bike with wings into a physics engine, and the gameplay loop being unlocking increasingly large airplane pieces, and designing - then flying - increasingly ridiculous airborne bicycles seems like a compelling idea, though less good for fitness.
The control surface being so limited - essentially one measurement which can either go up or down - is a really fun constraint for creativity.
I'd love to try a similar approach to gamifying exercise on a smart watch. That's an even more fun problem, because (when running, for example) you don't even have a constantly-present screen to communicate dopamine to the player with. Perhaps some watches wouldn't even be able to provide sound feedback? I've never owned one. It's a really interesting constraint for game design.
The interesting I noticed is that in the process of curating and organizing my photos, I stumble across memories and moments that I had completely forgot.
Development Challenge: The app does two interesting things underneath the hood... First, given an arbitrary number of thumbnails with different aspect ratios, it finds the best sized rectangle that completely fits all the photos. This allows me to recursively stitch groups together and always output a rectangular thumbnail for the group. Second, the pinch/zoom UI uses these nested thumbnails directly without slicing/dicing to fixed sized tiles.
https://shopspec.io
https://github.com/ReviewToolkits
https://github.com/devdanzin/fusil
https://lwn.net/Articles/1067234/
Meli: A spending tracker for Mac designed with privacy in mind. Turns credit card alert emails into spending insights and budgets. I was in security for 15+ years and I built it because I didn't want to use Plaid or other aggregators and wanted my financial data locally on my device and not in the cloud. Spreadsheets required a lot of work to maintain so I wanted some sort of automation. Connects directly to your email (reads only alerts from supported senders) and all parsing happens on your device.
Works with seven major US banks (AMEX, Capital One, Chase, Citi, Discover, Bank of America and Wells Fargo) plus PayPal and Venmo.
Does this list cover most of your spending? If not, which banks or cards would you need?
How do you get programmatic access to your data?
However, most banks do send alert emails when something is charged to a credit/debit card. Meli connects to your inbox read-only via OAuth (Google and Microsoft) or IMAP for other providers and reads those emails on your Mac.
Of course there are limitations to this approach, you only get access to the data that is available in those emails like merchant, amount, date and account.
Which banks are you interested in? Are you based in the US?
I thought I’d get where I wanted much faster, but it turns out I ran into quite a few challenges: it’s quick to build something simple, but it takes much longer to get a reader that’s robust and pleasant to use (extraction, parsing, security, infrastructure, etc.).
On the tech side, I’m using my usual stack, Laravel and Livewire, with some Python for extraction and Go for the email infrastructure. This month I’m working on the beta release, since several people around me were already using it, so I’d like to gather more feedback.
The serverless MariaDB is important because it can scale to zero, so you have dev sites that cost less than $0.50/mo.
Also added MCP and CLI tooling that exposes all of the infrastructure + WP CLI + DB so agents can develop WP sites... which is nice because you can still use the WP admin interface, while using AI to develop the site, debug it, change pages and posts, etc.
I'm very excited by adding interactivity. I have a new "knob" feature that pops out when you type {var}. It infers the column's type and values.
`SELECT * FROM hn_posts WHERE score>{score}` will grant you a slider that goes from 0 to 6015
Titanic Browser demo (no login, it's all WASM): https://app.kavla.dev/
Also been working on a "chat to data ingestion pipeline". Not sure where I'm going with this but I like the idea? https://github.com/aleda145/pompos
For the record, it means "to roll something out to a flat surface" in Swedish
I got overwhelmed by walls of text of parallel Claude Code sessions. This limited the sessions I could steer in parallel to maybe 5 or 6. I knew there had to be a better way, and I found one: A canvas per project. Work laid out as cards. An agent in a worktree per card. Chat condensed to WhatsApp-style back and forth, not a litany of tool calls. Result is usually a narrated demo video that I can comment on by voice to have the agent go another round. Remote-control from my phone. Remote workers to scale beyond my machine. Everyone I know who tried it has stuck with it so far.
https://github.com/sadilek/obeya
It’s been getting a decent number of sign ups despite me not really posting it anywhere or doing any marketing.
Ultimately a fun project, and useful for myself and a few friends, so I’m happy others are even interested in trying it.
I’ve started to regret using Phoenix LiveView, as it’s only deployed in one region, so high latency far from there. I’m exploring a SPA as a better alternative.
Otherwise, I’ve been playing around with a Lua engine to build feeds for websites that don’t support them, using AI to build the Lua as a once-off. It works reasonably well, but isn’t ready for prime time just yet.
https://github.com/swax/dub-anything
I use it to watch movies that have never been translated to English. A whole world of previously inaccessible media is opened up.
Then also HNowledge[1], a free iOS app for browsing the HN stories and comments I've upvoted. Made it to fit the way I browse HN, which is upvoting mainly what I find interesting and may want to come back to. The app lets you browse by date, domain or score, add notes and export. Apart from syncing your upvotes from HN, everything stays local without any tracking.
[1]: https://apps.apple.com/us/app/hnowledge/id6813417548
Wanes[2], an Apple Health logging interface for tracking how long symptoms last. I tend to have headaches quite often and the way Apple lets you enter that data is pretty limited, so I created an app which starts a Live Activity for one or more symptoms using widgets, Control Center or Siri. You can change the severity rating while it runs and stop it from wherever you are on your phone (Home Screen, Lock Screen, Dynamic Island, Siri).
[2]: https://apps.apple.com/us/app/wanes/id6811679904
It uses an allwinner T113-S3 SoC and the Quectel EC25 LTE module, and I've made sure voice calling works over LTE, at least on T-Mobile in the US. (this is tricky bc of carrier rules)
video and photos of MangoPi test setup and my board layout here: https://constcast.org/n1100c/
https://react.breaka.club
Because I can. Well, let's be honest: because LLMs can. There's nothing particularly novel about this. It's just wiring a bunch of libraries together. Certainly would have been time consuming, but not difficult, strictly speaking, since all the heavy lifting is done by the libraries utilised.
Likely going to migrate https://breaka.club HUDs and editor UI to React since Godot control nodes are by no means impossible to work with, but somewhat cumbersome.
EDIT: Oh, will be open source like a lot for Breaka Club's other libraries very soon. Still optimising.
It scans your codebase, turns it into permanent c4 & sequence diagrams and keeps them up to date on each commit. Then, instead of reviewing a 10k line coding PR, you review a 30 line change to sequence and C4 diagrams.
npm i -g @archme/cli && archme init to run it on your own computer, code never leaves your machine.
And Meltdown https://meltdown.merkoba.com/
Grasshopper is a tab manager for Firefox.
Meltdown is a an AI harness made in python and tkinter.
You can check screenshots of it and descriptions in the link.
Also I've been blogging a lot recently on a weight loss blog. I like asking Kimi for reviews on what I write.
I have a PAL-2 [1] (a modern KIM-1 re-creation) and started learning 6502 by entering programs from The First Book of KIM [2].
I wanted to be able to compile/link the assembly from the book so leaned on Claude to set up a 6502 dev environment for my Mac. I put everything needed (and instructions) on GitHub [3].
(Claude also has been helpful in re-formatting the 6502 code from The First Book of KIM into a more modern style compatible with modern assemblers. A few of these modernized programs I added to the GitHub repo above.)
As it happens, I am also fixing/formatting The First Book of KIM by recreating it in a modern page-layout app (Affinity Publisher). The PDF scans that are out there are not pretty, have OCR errors (and the book itself has mistakes). Hopefully I'll have a new version of the book/PDF out on the internet by the year's end.
[1] https://www.tindie.com/products/kim1/pal-2-a-mos-6502-powere...
[2] https://archive.org/details/The_First_Book_of_KIM/
[3] https://github.com/EngineersNeedArt/KIM-1_Development
The loop is stable and responsive when receiving from 17 to 29.9999 Mhz, but not much below that... Almost got it
Other than this, I'm playing with vibe-coding stuff I always wanted to do but never had the time/skills to do. For example, I built a TUI on top of beancount that is tailored to my needs and my business[2]. I also completely got rid of self-hosted plausible and umami analytics, and replaced them with vibe-coded built-in-rust (c) analytics tool that relies on both js-beacon and streaming of caddy access.log. Will probably publish about it in my blog soon, so subscribe to the RSS if you want.
Lastly, I'm messing with self-hosting. I completely ditched Github and Gitlab for self hosted forgejo + actions runner. I set up proper backup plan for everything I care (documents, immich library, my SaaS DB, and the forgejo repositories).
[1] https://justfaxonline.com
[2] https://kudmitry.com/articles/after-14-years-i-finally-built...
It's so amazing to have the ability to gather and parse publicly available scientific data and actually visualize it in real-time. It's one of those projects I'm learning a lot from and can keep going with (the universe is a big place :)). As a software developer for 20+ years, its been mind-blowing and often also frustrating to have Claude generate code that I deem good enough. About half of my time has actually been spent on refactoring, building quite some skills in guiding LLMs in the process. I guess that part still hasn't changed in a growing codebase.
I've been trying to improve agent knowledge over a codebase, and existing approaches weren't giving the result I wanted, and I wanted to keep it simple. In the end, I added a semantic layer on top of an LLM Wiki and it saved 30% of token usage in multiple benchmarks I ran. It also adds some nice features to deal with concurring agents updating the shared memory, to ensure all information there is reliable and fresh.
It's published to PyPi, so an "uv tool install wayfinder-memory" already does all the work, and it's compatible with almost all Harnesses.
I’m a software developer and watercolor painter so my goal is to use real world painting techniques and have the results feel and look real.
Modeling the physics of paper, brush bristles, pigments, water, color mixing, drying in real time using gpu.
Built in rust using wgpu.
Currently looking for iPad beta testers if anyone is interested (supports Apple Pencil or finger).
https://puddlewatercolor.com
To be honest, I started this because Postman and Insomnia became forced-corporate-cloud not usable for me anymore, and Bruno lacked a feature I considered fundamental for my workflow.
The feature is to remember the JSONPath expressions between requests, a feature that Bruno has implemented already.
But I created this thing and I'm using it daily since then. C++ and wxWidgets, it uses 40MB of RAM, and it is super fast, instead of the 400MB used by the above mentioned programs.
https://www.apikulture.com/
1. Broadcast (https://sendbroadcast.net) - Self hosted email marketing software. Been working on this since 2024 and positive revenue. Releasing v3 this week! :)
2. Render Screenshot (https://renderscreenshot.com) - Screenshot API (very much a scratch my own itch project)
3. Widgetful (https://widgetful.com) - A menubar app for managing multiple Claude Code agent sessions
It's a free job site with some resume tools bolted on. Lots of people from all over the world are using and that feels great. It has 2.7M+ jobs and automatically removes any job older than 7 days. We never allow reposts because, frankly, the job market is a nightmare and any job that has been up for 7+ days almost certainly has 1000+ applicants already. You're just feeding your resume into a wood chipper.
Come and try it. It's awesome.
- https://polygonjs.com/: a design tool to create interactive 3D configurators
- https://polyreprint.com/: a hosting platform for those configurators. See more enterprise-type examples on https://polygonjs.com/configurators
- https://polyreplay.com/: game made with said design tool
1. FormBeep(https://formbeep.com) at $38 MRR now - A form to Whatsapp tool. Stack is mainly CF workers with Clerk Auth.
2. SlashLink(https://slashlink.app) $0 - A simple pay as you go amazon link localizer. Great for affiliate folks and a cheaper alternative to Genius. Stack is same, CF workers with SQLite(D1)
3. StackStats(https://stackstats.app) $791 - A better advanced analytics app for Substack writers, fully local. I have been writing on Substack for a while now and their analytics suck!
Apart from that, running a travel agency I'd never be able to without agents (they manage my ads, social media, website content etc.). Started with boa.ba and it went quite well, now preparing dobosnia.com
It's a sandbox that you can embed ala the sqlite of VM's.
Link: https://github.com/smol-machines/smolvm
You can connect Claude desktop to databases using MCP servers, but Claude desktop is not great at displaying large volumes of data. A custom client can do a much better job.
https://reportr.com/
What does it do differently to https://getnao.io/ ?
Added a way to look at power/flops and new model requirements lately to try and estimate how many GPUs you need to run a certain model: https://flopper.io/tools/llm-gpu-calculator?model=llama-3-3-...
https://hn.orbitalchicken.com
There are more things I'd like to do with the graph concept, like linking two user's recipe trees if they were imported via the same URL. Really trying to to avoid features that I think would be cool that don't provide any functionality.
There are still a few features I'd like to implement that are more in the vein of nice to have in any recipe application, like nested recipes, temperature conversions, volumetric ingredients amounts to mass conversions, etc.
But I've been using it for new recipes and converting my old recipes I have written down in the margins to this format and I've been happy with how usable its been in these early stages.
[0] https://grainofsalt.app
The idea is to sandbox the most dangerous parts of it: The Document parser and renderer.
https://paulg.info/2026/11/04/quantum-weight-space/
github.com/patrickjh/ssa
https://cratediggr.com/
Artists make users jump through hoops to like, comment, repost etc.. before giving them access to the download - CrateDiggr highlights the free downloads in the Soundcloud UI and clicks through the download gates for users.
Been a fun passion project, check it out if you download tracks from Soundcloud!
It is New, Experimental, Modular, Declarative, Event Driven and Reactive. It builds cross-platform apps that can be compiled for the terminal (macOs, Windows, Linux), a standalone native window (macOS & Windows 11), or the browser (via WASM) from one unified code base.
Curious if this project will ever be 90% done.. ;)
https://www.phonique.com - An AVAudioEngine/PHASE Audio graph editor, it's easy to connect nodes together visually, then export the wired graph as Swift
https://github.com/maxmart/hoststray
Also it forces your browser to "flush" their DNS by killing the keep-alive connections that for example Chrome keeps open in background. And it can be controlled by your local coding agent through a simple API (which also serves the skill for it)
https://studio.flet.dev/run/bicQWc3Aul/
which is now at a working state for rabbet joinery boxes with a sliding lid, or omitting the lid:
https://community.carbide3d.com/uploads/default/original/3X/...
Next up is a hinged lid option and cleaning up the wireframe preview to look better/be more accurate.
The work mainly stems from a belief that Nvidia gpus' architecture are a bit subpar for machine learning and that tpu could be a viable alternative but are still rather undocumented.
Building the web-embeddable legal IDE for transactional AI-native law firms. Docx, PDF and Markdown in the browser with redlining, go-to definition and more. In Rust, of course.
site: https://notekpr.md
I'm kinda leaning towards vibe coding my own GUI coding agent. I currently use Pi and recently have been playing with Herdr and Paseo, but none of these feels right to me. These are amazing tools, but ai think we could do better. But I'm very hesitant - I know the only way to "succeed" in a project like this is to write (vibe code) hundreds of thousands of lines of code and it's kinda depressing.
https://github.com/tirrenotechnologies/tirreno
Live demo: https://play.tirreno.com
Also trying to replace own Duolingo usage with Fika Lingo (https://fikalingo.com) basic idea was to first repro the practice of matching words. There is now few other types of language / vocabulary practices. The practice generation works on-device with local LLM, you can give the topic yourself like "Words that relate to Hacker News". Downside at the moment is that English needs to be the counterpart of the practices. Cant guarantee good enough quality with something like germany <-> italy practices. I'm trying to do finetuning with LFM models to see if it would be feasible to improve the local practice generation.
Since the last time I posted on HN I've added 100s of cinemas, user features like being able to save search filters, movies, and specific showtimes. Also a ChatGPT plugin which allows for searching for movie showtimes using natural language directly inside of ChatGPT (https://chatgpt.com/plugins/plugin_asdk_app_6abfdc298c208191...).
Finally I've added AI search to the site as well, allowing for natural language search with queries like "English audio showtimes near Shinjuku tomorrow" and so on.
I would really appreciate feedback for things to add, things that are hard to understand, and anything else!
Now putting it to personal use and keeping an eye out for issues and rough bits to smooth over.
Built with -
NestJS, Sequelize + SQLite, OpenAPI, and Jest
React, TanStack, Shadcn, and Playwright
[1] - https://spill.rs
Along the way I ended up adding an internal agent framework to satisfy stuff like "log in, click the button, and then take a screenshot", out of which fell stuff like automatically generating Playwright tests from that. It even runs the Playwright test suite and Puppeteer test suites ( a bit faster, and for a lot less memory ). I kinda love this bit because I can layer in multiple models that will bounce control of the web session between each other, to do stuff like let a Typesafe model choose to escalate to Opus when it needs higher minded behavior. It's fun to watch it happen.
I've mentioned it a few times around here, I'll probably do a Show HN soon when I feel like it's ready for that. Hopefully soon.
[1] https://zshot-app.com/
the plot: Programming languages existed as an abstraction for humans. As writing code becomes cheaper, we can write it for each architecture as long as the specification is done well. The costs of running will matter more in relation. So don’t make your AI write rust, but assembly
and letting families hold off on screens for a few more years: https://readychime.com
I'm at the point where I'm finally actively using it to register my activity, it is straightforward to register tickets by sending them to a telegram bot, then, reconcile those by uploading my bank statement.
I'm trying to get the lowest possible usage friction so that I keep using it.
So far the better built pieces related to the data ingestion. Once this is good enough I'll build reports + anything coming to my head as a useful piece.
I'm leveraging AWS textract + bedrock + bda to parse bank statements. I expected this would have been easier but it is still a very complex problem which had required me to put many iterations (so far it works nicely with my own statements)
For ticket parsing, AWS textract + parsing rules has been enough.
I wrote a bit more on my blog https://alexitc.com/blog/2026-10-16-opensourcing-my-expense-...
Repo at https://github.com/AlexITC/expense-tracker/
Lately we've been working on access for AI agents. A lot of people want their own agent to do the repetitive part of time tracking. Control remains important, which is why we rebuilt our core this month to support change history and full undo/redo functionality. Feel free to try it with your own agent.
We built this after the timesheet tool we paid for raised its prices and moved some features that we used regularly behind more expensive plans. We had to add three new colleagues, and everything combined this would cost us $500 a year more for the same software.
We run Elixir and Phoenix LiveView on a Kubernetes cluster made of Talos instances.
We're still looking for feedback on the product, so please let us know what you think!
A simple EU static and app hosting service: PodBy [2]. It was launched last week, and is at €10 MRR.
Thank you for asking and for your suggestions!
[1]: https://uruky.com
[2]: https://podby.com
- hermitator, a tool to run multiple isolated AI agents for WhatsApp on a single machine, with idle agents using no RAM: https://www.encona.com/posts/hermitator/
- showntell, which allows your coding agent to explain things to you with a video that starts playing within a few seconds of you asking: https://www.encona.com/posts/showntell/
- a philosophy tutor (testing in private for now) that does 2:1 live philosophy lessons, each exploring a single topic (aimed at ~10yo kids). I've tested this successfully (i.e. two kids did an hour and they want to do it again), and am trying to make improvements.
- some other education-related things
- a jmail-style mailbox viewer: https://www.encona.com/posts/mailbox-viewer/
It combines nonograms with crosswords, requiring both logic and word knowledge to solve. I am still working on the interface and tutorial, as well as the puzzle difficulty levels.
been curious about the intersection between observability, SDKs, and product analytics. i couldn't find anything around SDK observability so built this little tool. curious what yall think!
i assume at bigger orgs there is some internal handrolled tool like this that ingests datadog/sentry/whatever info...if anyone plays around in this space and has feedback id love it!
I'm a product designer by trade and learnt to vibe code on Claude about a month ago now. I was weaning my baby and he developed a rash and haven't been able to work out what the pattern was causing it. So I created an app for weaning parents to log the food, add any reactions so when talking to their GP they can start to see a pattern and all advice, like nutrients, food variety, how to serve food all coming from the NHS and FSA.
Currently got 12 mums using the app and would love any feedback to make it even better! <3
https://apps.apple.com/gb/app/niblet-weaning-diary/id6822680...
https://getniblet.app
Here's one about speculative decoding (and how it speeds up AI inference): https://www.youtube.com/watch?v=cGnIKjGIwjA
Most planners treat your schedule as fixed, so the moment a meeting runs long or a task takes twice as long, the plan is wrong and you either fix it by hand or give up on it. In Ember, if you move a task or change how long it takes, everything after it shifts later, and it flows around fixed things like meetings. It's been super helpful and I have been adding new features ever since.
Right now I'm working on a native Kotlin app to replace the current one is web + Capacitor and exploring full duplex voice models so I can add or move things around the app while having a conversation with an agent.
homepage: https://emberflow.space demo: https://demo.emberflow.space/?ref=hn
I've got 40k+ photos on my iphone, and wanted a better way to interact with them. The basic idea I had was to virtually stitch all the photos together into a gigapixel image, and navigate that image with pinch and zoom. On import, the app hierarchically groups photos based on date and similarity. The user can go back and edit the groups, change the layout, and adjust the size of photos if they choose.
The interesting I noticed is that in the process of curating and organizing my photos, I stumble across memories and moments that I had completely forgot.
Development Challenge: The app does two interesting things underneath the hood... First, given an arbitrary number of thumbnails with different aspect ratios, it finds the best sized rectangle that completely fits all the photos. This allows me to recursively stitch groups together and always output a rectangular thumbnail for the group. Second, the pinch/zoom UI uses these nested thumbnails directly without slicing/dicing to fixed sized tiles.
I'm building a polyglot web framework in 15 languages, with C# and Lean4 as the reference implementations. Structural and behavioral audits using deterministic tools check that every port stays consistent with the reference. It's about 1.8M lines now. I have not reviewed the code/neither have i verified the behavior tests except at the time of giving requirement and never seen some of the language code at all.
The question is how far can we produce "working code → correct code [Possible] → good code [Unknown]" with deterministic audits correcting structure, ai mistakes and behavior validation.
So far the quality looks acceptable with lots of correction of AI output done without looking, but it isn't fully verified.
Line counts per language, with changes since August: https://claude.ai/artifact/5f1m1pTM7tZ4boQzwsoAG7
A bit like iodine in the basic idea, but the execution and network topologies you build with it are different. It's more of a building block; you get your DNS channels represented as UDP endpoints, and take it from there, ie you can run Wireguard over it, or SSH or a SOCKS5 proxy (using quicmux, so over QUIC), or run Mosh straight over DNS if you're crafty.
It has an Android app too! I tried to submit it to F-Droid but they're so swamped in spammy or low-quality apps that they're not taking any brand new apps :-( A sad state of affairs, that.
It's been a fun project to work on since it's been a pain point for myself on many projects, so I'm mostly just trying to solve my own set of problems and hoping other people find it useful too.
[1] https://sendery.co
I was trying to solve the problem I had with time-blocking, where your day ends up derailed and you give up on it because it takes too long to replan. I have solved with the idea of a cascading calendar. If I move a task down or change its duration it pushes other tasks later in the day.
Since then its kinda grown into my full productivity system for logging my thoughts and planning and documenting my life
https://emberflow.space
You can demo here for free https://demo.emberflow.space/?ref=hn
https://github.com/gabrilend/ai-stuff
Not really sure what to call it, Linux kernel ish implemented in JS so you can run it in the browser. Each program runs in WASM. Everything bootstrapped from a C/Cpp compiler compiled from source with a new target. Also kinda turned into Docker for WASM.
Recommended for people that feel a strong urge to run a browser inside a browser.
https://slopshare.dev/v0.1.5/wine/
Also quite useful to run coding agents sandboxed inside the browser, with HTTP policies and "backend managed skills/extensions". Compiles Codex from source (has a rust compiler), runs pi agent and CC (the old node one) on a custom javascript runtime on top of quickjs compiled inside the environment.
Everything is static JS (or the robots are lying to me)
My new favourite shader running realtime over your camera input. Has gotten great feedback from kids as well :)
Knowers of plasma pong might see slight resemblance.
Send your beacon to a group, and everyone gets a gentle ping. The first person who answers gets connected for a 1-1 call, and for everyone else the signal drops silently. It does not log any missed calls so there’s neither pressure to answer nor pressure to return calls.
Overall it works really well! It’s best if you're able to join with 4+ friends/family you’d like to chat with more. I have dozens of connections in the app now and it feels like absolute magic to me. I tap the button to send my beacon to everyone and get a random friend on the line.
https://trybeacon.chat/
I’m looking for ways to fund printing and distribute the book. If anyone has done something similar, I’d love to learn from your experience!
The main feature is, that I can use different subscriptions/API keys (Claude and Codex - v0.1.4) and trackers (GitHub, Linear and Huly - v0.1.4) to make Even Keel start an agent for every single issue, task by task, one at a time per folder. There is also a parallel feature, so I can run more than 1 project at a time.
Once upon a time, when codex agent was integrated into the app, I wrote a ticket to make codex his own hook. It reused a hook that blocked its own git merge, if tests did not pass. I started calling it "A new agent writes its own cage"... Right after, it turned out it doesn't listen to the hook and it fixed it
Version 0.1.4 is already shipped, but v0.2 is coming soon with more supported trackers, agents and new features.
https://github.com/NamerekT/even-keel
My first idea is to require the user to have more than one disk for the system needs and the system should know all of the rest data devices user owns. My second idea is to treat different files as having different priority which impose multiple copies. My third idea is to make the software accessible, and the last idea is to make encrypting of all the devices easy and with no ties to any commertial entities. I would like to have a discussion about how to not lose anything digital.
It started with building pi in the cloud - with pure cloud flare workers. I have it working now, and can spin up - the wrangler agent can fully run it as long as you pass it an inference key, including a file view, a chatgpt like experience, and ability to build and host custom apps. Can use alarms to wake up to do plain language tasks, etc.
Trying to figure out how to package this up, and where this would even be useful for other apps.
Also revived old project - https://aminosplore.metacursion.com - the idea was to compare what essential amino acids each food contains using spider web chart. It’s been dead for a while due to USDA API changes so I revived, added macros and enabled to build foods into actual recipes.
Finally have been improving elsrc.com - an easy and scalable way to use ai image generators. If you need hundreds of images you can just request them by url. Supports versioning, batching, regional and national styling based on viewers IP and few other neat features!
I like sports! I like data! Combine!
This lets you see what the games on today are most similar to in the history of the sport. Click into one and see.
Also, the Sports Zodiac.
Like farmers of the olden years and the stars, we kinda set our clocks by sports these days. But what does that look like?
'Play the Years' to see a good timeline of how the years go by.
The war years, the lockouts, and the pandemic are just really easy to see in the data and quite stunning really
It’s a one-to-many phone call that only one person can answer. Send a beacon to a group, and everyone gets pinged at the same time. The first person who answers gets connected for a 1-1 call, and for everyone else the signal drops silently. It does not log any missed calls so there’s neither pressure to answer nor pressure to return calls.
Overall it works really well! It works best if you're able to join with at 4+ friends/family you’d like to chat with more. I have dozens of connections in the app now and it feels like magic to me. I press the button and get a random friend!
https://trybeacon.chat/
Not quite ready to open up sign-ups, but you can get an idea of the site here: https://linkedl.ist/@kale
The idea is that the source code is annotated with special rules which configure how parts of the code (also configs and docs) depend on each other. This can be done either directly in the source code comments or in a config file blockwatch.toml
This tool is very useful for large codebases that e.g. share constants across programming languages and configs, have docs that should strictly reflect the current state of the code, etc
It uses tree-sitter to parse the source code, 35+ languages are supported.
The project is actively developed, your feedback is welcomed! Thanks!
I always got annoyed by change this h1 to h2, add tracking to share button, why can’t the customer play the video, the styles don’t match figma.
If you are engineer who also gets distracted I would love to hear your opinion and feedback. It is a bit hard to put it all together as every department has his own needs and opinion about the website.
My email kamran.elchuzade@gmail.com
Thesoapbox.email - a social experiment email lottery to see what people do when suddenly given enormous reach. Every day a new subscriber is drawn to email the whole list. It’s hit 10,000 views in a couple of weeks which is my most viral project so far.
The second - a personal AI learning platform built on recursive self improvement and tailored content to push my boundaries. I’m a lifelong learner and this unifies everything for me. (Currently not public)
and this weekend I'll finally launch a tool for guitarists and their friends https://kantamigo.com/
Probably the most interesting thing about it at the moment is the editor. The world, and most things in it, are represented as collections of SDFs. More accurately, they're density fields, but, potato-tomato. One neat thing about representing the world as SDFs is they've been thoroughly researched and documented by a guy named Inigo Quilez, and they have a lot of nice mathematical properties. For example, you can do a smooth union of arbitrary SDFs to get nice rounded contours where shapes join.
The world generator and editor operate almost entirely on the GPU. The terrain and editor pipeline is a three stage process. The first stage is user-configurable; users write an arbitrary 'shaping' shader in GLSL. The second stage is a fused stage; it does a convolution on the output of the first stage to compute a normal, runs a second user-configurable 'decoration' function, and finally applies edits. Edits are collections of SDFs that are 'compiled' from a graph structure on the CPU into a bytecode, and interpreted in the shader. The last stage runs another convolution to compute the final normal, then packs an occupancy bit, color, and normal down into a 32-bit representation, suitable for transfer back to the CPU. I expect I will expand this to include additional user-level bits for encoding biome type and GPU-driven spawning.
The entire shader pipeline is one fused 'ubershader' .. everything is bound at startup and the shader is dispatched three times, incrementing a 'Stage' uniform on each dispatch. This strategy conceptually makes the driver overhead nearly zero, and the profile agrees.
The final step in the process is giving the CPU a chance to spawn additional entities and structures after generation. This is not completely fleshed out, but the bones are there. The idea is that the results of the GPU work are passed back to the CPU, which can run arbitrary callbacks to spawn additional edits and entities. Once spawning is complete, the result is meshed using a fast binary meshing algorithm and uploaded to the GPU. I keep 16 bits of data per voxel from the GPU work, 1 bit of occupancy, and 15 bits of normal for collision response, which works out to half a megabyte per chunk. At the moment I only keep chunks which contain a piece of the world surface, and discard the rest. Surface chunks are dense, and may be compressed in the future, but keeping this amount of data scales to world sizes greater than 1B voxels cubed, which seems good enough for now.
In terms of raw performance, GPUs are ridiculous. My 3080ti mobile gets somewhere on the order of 500B-1T perlin samples per second with a naive perlin implementation. Compared to FastNoise2, that's about 1000x more, which sounds wrong since my GPU only has 35x more capacity, but, that's what my initial profiling said. That number is the marketing number; the real throughput is significantly less due to multi-stage nature of the system, but, there's a lot of headroom for improvement. For example, dropping the backing textures from 32bit to 16bit roughly quadruples performance. Since we only really care about the surface where values are near zero, I believe using 8 or 16 bit is a viable option, though this claim will require a fair amount of study to validate.
All in all, I'm pleased with how this is all working out. My original goal was to write a highly flexible system giving users both easy-to-use tools and granular control over the final output. I think I've done a fairly good job at that, and the fact it's also super fast is the cherry on top.
I'll do some shameless self-promotion and leave some links here for anyone interested in looking at the engine, language code, or some pretty pictures.
https://github.com/scallyw4g/bonsai
https://github.com/scallyw4g/poof
I just started using AI a few months ago. My smoke-test was "can it write a simple memory allocator". Until recently, every time I tried to get an AI to write an arena or simple buddy-style heap allocator, they just vomited all over themselves. To my surprise, the last time I tried, the AI absolutely crushed it; it unearthed tricks I didn't even know about in one shot. Since then, I've been using AIs to do the boring, rote work that effectively just boils down to me typing in something that's in my head. For that, they're great. I typically sketch out the thing I want it to make, then tell it to fill in the blanks. Sometimes they still go off the rails, but all in all it's a useful tool.
Where I've found they still struggle is making any architectural decisions, or where performance matters.. even a little bit. They often make super braindead decisions when it comes to performance for no discernable reason. They'll course correct when you tell them to, but I've had to keep a close eye on the output and reign it in frequently. Asking an AI to design a system from the API level hasn't ever returned a good result for me.. I always do that work.
All in all, I feel like using AI will probably make me a better programmer in the long run, contrary to popular belief. They accelerate the boring stuff I already know how to do, which leaves me with more time to work on the edge of my skill level where I grow as an engineer. Guess we'll see in 5 years.
Every puzzle is run through a verifier first to make sure it is solvable without any guessing, just by pure logic, so there is always a next step to find, which is what makes it feel so rewarding. There's a hint system to help and it works really well if I do say so myself . No account needed to play.
From what I've been hearing so far, the people who have checked it out are enjoying it and can't wait for tomorrow's puzzle, so that's a good sign. I'd love to hear how the difficulty feels, and whether the pictures read clearly.
https://marketyour.games
https://prophitt.me/writing/a-modern-web-app-in-c
Right now I’m mainly validating whether people would actually want something like this. If there’s enough interest, I’ll run a small closed beta for Android.
Join the waitlist for beta access: https://forms.gle/XWXmK8HUSmAtfms29
So far it's helpful for messages between my own AI's (chatgpt and claude etc), and for having my AI trade tech tips with a friend's AI or talk to my wife's AI. My AI also got its first message from a strangers AI from france which has also been really interesting. I read all messages first and approve before sending anything, but otherwise they're just talking about whatever they want.
It's a traditional systems language in the sense that you have fine control over how resources are used, but it's abstract in the sense that you have to come to an agreement with the compiler about your intent around the resources, rather than directly declaring the machinery.
It's a distributed language in the sense that there is a higher level structural abstraction in the language that situates computations of the program within a recursive graph structure. Part of that structure incorporates message broker semantics. This gives the compiler a unified view of programs across arbitrary process boundaries. The structural abstractions also enable an effects+formal verification syntax within the language, which allows programmers to codify architectural decisions that are then checked by the compiler.
The compiler itself is infrastructure aware, and the highest layer of functionality it handles is configuring the environment into which applications deploy. The single language handles everything from individual memory allocations to distributed fleet deployments.
The language is close to v1, and I'm currently building a fully distributed inference-as-a-service application with it. Arbitrary hardware/supply matched against arbitrary callers/demand in real time.
https://github.com/hale-lang/hale https://github.com/hale-lang/voice
You write or dictate an entry, and a small model (LFM2.5 1.2B) asks you a question to reflect further. It turns out 1.2B models are decent at asking questions, so I leaned into that. An embedding model finds older entries about the same thing and shows bits of them under the new one.
Most people using it are on budget phones running the models on the CPU, where the app provides good experience.
https://tinta.page
It has if nothing else been very useful for me (it has correctly deduced that I would get something out of borrowing Grief is the Thing With Feathers, the Dispossessed, and Cosmicomics, none of which I would have thought of reading). Plus it has been a fun opportunity to play around with LLMs as an interface and a research and reverse engineering tool; as well as Jev for non-user facing quality work like banning crap openrouter providers based on their output before putting it in front of a user.
Most recently I’ve been dropping off flyers for it at a few local libraries in my area.
Also have some other cool browser-based 3d game projects listed here: https://brynnbateman.com/
Also have some other cool browser-based 3d game projects listed here: https://brynnbateman.com/
including an online version of Shenmue: https://newyokosuka.com/
Storyline isn't available yet, but about 75% done with it
* files in organised trees/graphs generated using LSPs and symbol references instead of linear file list * keyboard shortcuts for everything from navigation, pan, zoom, comment, note, submit... * mostly github compatible with native features * LSP powered symbol resolution for inspecting definitions and references * LLM powered reviews supported * Inline LLM question answering
Repo: https://github.com/viciousnemesis/housetrained
Puppy: https://imgur.com/piDE0gq
[1] https://apps.apple.com/us/app/reflect-track-anything/id64638...
https://porchlit.app/
I built the original for myself because I wanted to listen to acoup and other history blogs while sleeping. And in the spirit of getting better at finishing projects decided make it public even tho it’s nothing really unique or special.
Still lots to polish especially in the text/copy and ui but the core features should be working.
Looking to pilot for free with small qualified teams (early-stage startups with real/immediate need).
hello@llmsolutions.ai
It uses Apple Intelligence to help brainstorm domain name ideas. I’m interested in seeing what small local models can do as they catch up to larger ones, this is my first iOS app so I started with something simple that scratches my own itch.
Then I'm working on improving my two open-source apps, Portfox [2] and Localfox [3]. Portfox shows ports for running services and dev servers, Localfox provides them with .localhost domains with TLS and Cloudflare tunnels.
[1] https://wishfox.app [2] https://github.com/xergic/portfox [3] https://github.com/xergic/localfox
https://github.com/InditexTech/kumoss
Basically allow developers to request any infrastructure need in a web portal and automatically that request is fulfilled following the organization standards the platform team defined.
Any feedback or comment is more than welcomed!
Any feedbac
Added an audio player this past month for all fellow file hoarders out there.
Also, you can now choose if you want Gusco to live in your terminal or run as a native-yet-texty app.
Was pleasantly surprised how well turning a Bubble Tea TUI app into a GUI-ish bundle works with Gio UI.
https://github.com/devlensio/devlensOSS
Jokes aside (although I'm not entirely joking about my mental health), my main project is https://auradistill.com: First-principles thinking for your AI agents.
It's not just about giving your personal AI agents "memory". It's about encoding memories in a way that can handle ambiguity and contradictions.
That project alone has got me completely hooked on building for and with LLMs.
I've got other projects too, but this is probably the one I've poured the most time and energy into.
switching from iOS, I wasn't happy with the existing podcast apps, so I built this mostly for myself. I've been using it for the last six months and finally pushed a public beta.
My day to day is BE so I used the project to learn Kotlin and Jetpack Compose. Everything is strictly native using Room and Media3/ExoPlayer. I wanted a nice looking (subjective), privacy-focused app without any tracking or ads.
I'm considering open sourcing it soon.
https://play.google.com/store/apps/details?id=com.enyodigita...
Taking a break from features to focus on refining content filters and expunging sites that slipped through. Looking at you, .ai TLD.
https://outerweb.org
and many other things
Features:
- No paywalls, bandwidth limiting, or crippled behavior. No login required.
- Single go binary, deployed via Docker.
- Reaches your home server from anywhere with no port forwarding, no VPN and no paywall, even behind carrier-grade NAT. No Tailscale required.
- Cast to another device. You can use any logged in device as a playback screen.
- Watch together, with voice support: Play synced video with friends, and also chat voice. You can also mix this with casting to another device.
- Dialogue boost and loudness levelling for every film.
- Real adaptive quality, tuned for a home's upload.
- Hardware transcoding, full quality in every supported browser.
- A library that explains itself: what the scanner saw, why each title matched, guesses marked for you to check, and a disk that's gone missing shown as missing instead of its titles silently vanishing.
- Your data stays yours: The database is a rebuildable index, metadata corrections can be written back as NFO files that Kodi and Jellyfin read, and an automatic backup keeps what can't be re-derived (watch state, preferences, shares).
Security is a primary concern:
With the ever increasing pace of hacks and CVEs and AI-assisted intrusion, you simply can't be too careful anymore.
- The core (which talks to the network, handles the API, users and the database) never opens a media file or runs FFmpeg.
- A separate media node does every file read and every transcode. It has no network access and talks to the core only over Unix sockets.
- FFmpeg runs under Landlock and seccomp with no path to the file and no socket. All it can do is read and write streams.
- Small dependency tree on an allowlist, pure Go with no cgo, secrets sealed at rest and never logged.
- The library can be mounted read-only, so even a full compromise can't encrypt or delete your media.
- Cryptographically signed chain for servers and releases.
https://orpheum-app.com/
https://smashcut.games
I didn't want to switch/refresh bunch of tabs. That's why I made this one.
not screenshots. so you can still interact with the pieces like in normal web browsers.
But apart from that, I just updated my website's theme to something I wanted to do for quite a while now – a fantasy desktop! If you'd like to check it out: https://0xff.nu
I'm also trying to just learn new things with the time I was refunded by being unemployed.
Also building calculators for things engineers usually calculate in a spreadsheet at 2am. you can see some on the site(https://buildopsy.com/calculators.html)
[1]https://www.makonea.com/en-US
Client posts request and masters respond with their offers.
What you think about that guys?
https://typequicker.com
Good news though, these days Claude will just do it for me!
I’ve been working on setting up a home media server, using headless Debian on a pi - I’ve got it set up to stream music and videos, hold all my ROMs, I’ve got it automatically standardizing file names and handling tagging and subtitles - all attached to a big 6-drive enclosure that I just stuffed all my random old HDDs into.
Even though I haven’t written more than a dozen lines of code or commands for the whole thing, I am pretty fuckin pleased that now I can just wave a magic wand and say “I want all my video game soundtracks filed under the game name rather than the artist” and Claude uh finds a way to grant my wish. I spend about an hour each night tweaking it and it’s so easy to customize to be exactly what I need.
We’ve just incorporated in Berlin together with two great co-founding colleagues. Honestly never felt more excited.
Though it is a B2B tool, check it out: https://esara.ai
The terrain and editor pipeline is a three stage process. The first stage is user-configurable; users write an arbitrary 'shaping' shader in GLSL. The second stage is a fused stage; it does a convolution on the output of the first stage to compute a normal, runs a second user-configurable 'decoration' function, and finally applies edits. Edits are collections of SDFs that are 'compiled' from a graph structure on the CPU into a bytecode, and interpreted in the shader. The last stage runs another convolution to compute the final normal, then packs an occupancy bit, color, and normal down into a 32-bit representation, suitable for transfer back to the CPU. I expect I will expand this to include additional user-level bits for encoding biome type and GPU-driven spawning.
The entire shader pipeline is one fused 'ubershader' .. everything is bound at startup and the shader is dispatched three times, incrementing a 'Stage' uniform on each dispatch. This strategy conceptually makes the driver overhead nearly zero, and the profile agrees.
The final step in the process is giving the CPU a chance to spawn additional entities and structures after generation. This is not completely fleshed out, but the bones are there. The idea is that the results of the GPU work are passed back to the CPU, which can run arbitrary callbacks to spawn additional edits and entities. Once spawning is complete, the result is meshed using a fast binary meshing algorithm and uploaded to the GPU. I keep 16 bits of data per voxel from the GPU work, 1 bit of occupancy, and 15 bits of normal for collision response, which works out to half a megabyte per chunk. At the moment I only keep chunks which contain a piece of the world surface, and discard the rest. Surface chunks are dense, and may be compressed in the future, but keeping this amount of data scales to world sizes greater than 1B voxels cubed, which seems good enough for now.
In terms of raw performance, GPUs are ridiculous. My 3080ti mobile gets somewhere on the order of 500B-1T perlin samples per second with a naive perlin implementation. Compared to FastNoise2, that's about 1000x more, which sounds wrong since my GPU only has 35x more capacity, but, that's what my initial profiling said. That number is the marketing number; the real throughput is significantly less due to multi-stage nature of the system, but, there's a lot of headroom for improvement. For example, dropping the backing textures from 32bit to 16bit roughly quadruples performance. Since we only really care about the surface where values are near zero, I believe using 8 or 16 bit is a viable option, though this claim will require a fair amount of study to validate.
All in all, I'm pleased with how this is all working out. My original goal was to write a highly flexible system giving users both easy-to-use tools and granular control over the final output. I think I've done a fairly good job at that, and the fact it's also super fast is the cherry on top.
Most of the essential features are implemented so far, but some (mandatory) features are lacking (xattrs support). My goal is to get to a full ext4 implementation that also supports ext2 and ext3 read/write.
The initial motivation was to have a pet project with a use-case for Rust, as I wanted to improve my Rust coding skills, after I did some work on a game server implementation in Rust earlier [1]. My goal is to have an implementation where I have digested every line mentally, which is something that pretty much excludes slop-like use of AI for large scale code generation. And of course I want to really trust the code that touches my files via raw hard disk access. Hence, I use AI mainly for unit tests and (scoped) code rewriting when my code gets too verbose and cumbersome.
In the long run I'd want to add fuzzing to the implementation, so if anyone has experience on this, I'd be interested in getting some pointers.
[1] https://github.com/MordorWide/EANation
https://apps.apple.com/app/id1464477788
...according to my own benchmarks (yeah, I know how that sounds).
Next, I'm planning to submit it to independent third party benchmarks, hoping to find out whether my results are accurate... or not :)
I've also just finished a CSS framework: https://iriscss.com/
Now I'm working on a new backend programming language that runs on the Erlang BEAM, along with a few web applications built using this tech stack.
https://ur.dev/en/
(3 days ago /25p/56c) https://news.ycombinator.com/item?id=50389501
(7 days ago /23p/77c) https://news.ycombinator.com/item?id=50328340
Will merge the others.
Edit: ok, that's done and I did my best to normalize the timestamps for the posts from the older threads, but it's a bit messy. If you see weird things that's why.
Take a look at FFmpeg as an example. It's a complex tool that can do most things audio/video. A power user can do magical things, but they still provide simple commands for the most basic use-cases like conversion or trimming.
https://www.clipcat.com/blog/how-to-use-ffmpeg-a-beginners-g...
1. Provide sensible defaults 2. Study user journeys and write simple manuals for them. 3. Ask friends or internet communities for feedback.
Don't get discouraged, you can do it!
Accept online card payments from just 0.79% (UK) and earn 0.89% on all card spend [1], or get paid out to your bank.
I'm building tuppence to be available in the US, and 20 EU countries.
I'm getting a Point of Sale (POS) manufactured in the UK [2] for those who want to accept payments in person with card readers.
My goal is to help people sell their goods and make a living without paying over the odds for the privilege.
I charge a competitive fee for payment processing, give money back on card spend (with no cashback caps/limits) and do not charge saas fees or lock people into contracts.
Tuppences currently allows people to sell online with a store builder, expense management, payment links, direct debits & bookings - essentially Tuppence is a 'business' layer on top of what you get from the likes of Stripe, Adyen and Checkout.com. There's also have a 'connect' platform, so you can roll this out to your marketplace business.
Next on my roadmap is ticketing and 'treasury'. Treasury i'm personally excited about as you'll be able to earn 4% yield on your earnings, FDIC Insurance (FSCS in the UK), multi-currency banking in fiat balances such as USD, EUR, and GBP, or stablecoin balances like OUSD and USDC, and have bank account numbers to send or receive funds.
---
It's brand new, i'm still testing payment -> payout flows. These take a little time to verify on live payment rails.
Happy to give some fee-free payment processing for those who reach out and want to be early adopters, I can be reached at ollie@tuppence.ai, or comment on thread.
---
Tech Stack TLDR:
* TypeScript monorepo covering the API, dashboard, docs, mobile app and SDKs. * The API runs on Ts.ED over Express with TypeORM and Postgres. * Background jobs use pg-boss, so the queue lives in Postgres and there’s no Redis or separate broker. * The dashboard, till and storefront are React + Vite + Tailwind, the docs and help centre are Astro, and the mobile app is React Native/Expo. We ship SDKs for Node, Python, iOS, Android, React Native and WordPress, plus an MCP server and agent tools. * Everything runs on Google Cloud Run and Cloud SQL.
References:
[1] Card issuing: https://tuppence.ai/cards [2] Point of sale: https://tuppence.ai/pos/ridge/
Guaranteed results are achieved using the state of the art adaptive TDEE (total calorie burn of the day) tracking technique we've developed. It's sensorless, so no watch or other wearable is needed. It's way more accurate than those wearables (data included on the website), and our group has 2,000+ members ( https://macrocodex.app/community/ ). You can also read Play Store reviews.
All this completely free, no subscription or ads: https://macrocodex.app/
MacroCodex also works on both Android and iOS and any platform where PWA app can run.
Other other 2 apps are Android only (for now):
If you are looking for a workout app with Alpha Progression or MacroFactor Workout-like “Smart Progression,” or an app where you can run any program like Boostcamp but ad- and subscription-free, check out the Symbiote workout app (beta) https://symbiote-studio.macrocodex.app/
IF you are looking for AI photo Calorie tracking with lots of features which make it very fast to log food on daily basis checkout CalorieCodex (beta): https://caloriecodex.macrocodex.app/
All the apps have been under active development for a few months now.
This is in both https://news.ycombinator.com/newsguidelines.html and https://news.ycombinator.com/newsfaq.html.